Author Archives: Kimberly Samra

Mitigating prompt injection attacks with a layered defense strategy

Posted by Google GenAI Security Team

With the rapid adoption of generative AI, a new wave of threats is emerging across the industry with the aim of manipulating the AI systems themselves. One such emerging attack vector is indirect prompt injections. Unlike direct prompt injections, where an attacker directly inputs malicious commands into a prompt, indirect prompt injections involve hidden malicious instructions within external data sources. These may include emails, documents, or calendar invites that instruct AI to exfiltrate user data or execute other rogue actions. As more governments, businesses, and individuals adopt generative AI to get more done, this subtle yet potentially potent attack becomes increasingly pertinent across the industry, demanding immediate attention and robust security measures.


At Google, our teams have a longstanding precedent of investing in a defense-in-depth strategy, including robust evaluation, threat analysis, AI security best practices, AI red-teaming, adversarial training, and model hardening for generative AI tools. This approach enables safer adoption of Gemini in Google Workspace and the Gemini app (we refer to both in this blog as “Gemini” for simplicity). Below we describe our prompt injection mitigation product strategy based on extensive research, development, and deployment of improved security mitigations.


A layered security approach

Google has taken a layered security approach introducing security measures designed for each stage of the prompt lifecycle. From Gemini 2.5 model hardening, to purpose-built machine learning (ML) models detecting malicious instructions, to system-level safeguards, we are meaningfully elevating the difficulty, expense, and complexity faced by an attacker. This approach compels adversaries to resort to methods that are either more easily identified or demand greater resources. 


Our model training with adversarial data significantly enhanced our defenses against indirect prompt injection attacks in Gemini 2.5 models (technical details). This inherent model resilience is augmented with additional defenses that we built directly into Gemini, including: 


  1. Prompt injection content classifiers

  2. Security thought reinforcement

  3. Markdown sanitization and suspicious URL redaction

  4. User confirmation framework

  5. End-user security mitigation notifications


This layered approach to our security strategy strengthens the overall security framework for Gemini – throughout the prompt lifecycle and across diverse attack techniques.


1. Prompt injection content classifiers


Through collaboration with leading AI security researchers via Google's AI Vulnerability Reward Program (VRP), we've curated one of the world’s most advanced catalogs of generative AI vulnerabilities and adversarial data. Utilizing this resource, we built and are in the process of rolling out proprietary machine learning models that can detect malicious prompts and instructions within various formats, such as emails and files, drawing from real-world examples. Consequently, when users query Workspace data with Gemini, the content classifiers filter out harmful data containing malicious instructions, helping to ensure a secure end-to-end user experience by retaining only safe content. For example, if a user receives an email in Gmail that includes malicious instructions, our content classifiers help to detect and disregard malicious instructions, then generate a safe response for the user. This is in addition to built-in defenses in Gmail that automatically block more than 99.9% of spam, phishing attempts, and malware.


A diagram of Gemini’s actions based on the detection of the malicious instructions by content classifiers.


2. Security thought reinforcement


This technique adds targeted security instructions surrounding the prompt content to remind the large language model (LLM) to perform the user-directed task and ignore any adversarial instructions that could be present in the content. With this approach, we steer the LLM to stay focused on the task and ignore harmful or malicious requests added by a threat actor to execute indirect prompt injection attacks.

A diagram of Gemini’s actions based on additional protection provided by the security thought reinforcement technique. 


3. Markdown sanitization and suspicious URL redaction 


Our markdown sanitizer identifies external image URLs and will not render them, making the “EchoLeak” 0-click image rendering exfiltration vulnerability not applicable to Gemini. From there, a key protection against prompt injection and data exfiltration attacks occurs at the URL level. With external data containing dynamic URLs, users may encounter unknown risks as these URLs may be designed for indirect prompt injections and data exfiltration attacks. Malicious instructions executed on a user's behalf may also generate harmful URLs. With Gemini, our defense system includes suspicious URL detection based on Google Safe Browsing to differentiate between safe and unsafe links, providing a secure experience by helping to prevent URL-based attacks. For example, if a document contains malicious URLs and a user is summarizing the content with Gemini, the suspicious URLs will be redacted in Gemini’s response. 


Gemini in Gmail provides a summary of an email thread. In the summary, there is an unsafe URL. That URL is redacted in the response and is replaced with the text “suspicious link removed”. 


4. User confirmation framework


Gemini also features a contextual user confirmation system. This framework enables Gemini to require user confirmation for certain actions, also known as “Human-In-The-Loop” (HITL), using these responses to bolster security and streamline the user experience. For example, potentially risky operations like deleting a calendar event may trigger an explicit user confirmation request, thereby helping to prevent undetected or immediate execution of the operation.


The Gemini app with instructions to delete all events on Saturday. Gemini responds with the events found on Google Calendar and asks the user to confirm this action.


5. End-user security mitigation notifications


A key aspect to keeping our users safe is sharing details on attacks that we’ve stopped so users can watch out for similar attacks in the future. To that end, when security issues are mitigated with our built-in defenses, end users are provided with contextual information allowing them to learn more via dedicated help center articles. For example, if Gemini summarizes a file containing malicious instructions and one of Google’s prompt injection defenses mitigates the situation, a security notification with a “Learn more” link will be displayed for the user. Users are encouraged to become more familiar with our prompt injection defenses by reading the Help Center article. 


Gemini in Docs with instructions to provide a summary of a file. Suspicious content was detected and a response was not provided. There is a yellow security notification banner for the user and a statement that Gemini’s response has been removed, with a “Learn more” link to a relevant Help Center article.

Moving forward


Our comprehensive prompt injection security strategy strengthens the overall security framework for Gemini. Beyond the techniques described above, it also involves rigorous testing through manual and automated red teams, generative AI security BugSWAT events, strong security standards like our Secure AI Framework (SAIF), and partnerships with both external researchers via the Google AI Vulnerability Reward Program (VRP) and industry peers via the Coalition for Secure AI (CoSAI). Our commitment to trust includes collaboration with the security community to responsibly disclose AI security vulnerabilities, share our latest threat intelligence on ways we see bad actors trying to leverage AI, and offering insights into our work to build stronger prompt injection defenses. 


Working closely with industry partners is crucial to building stronger protections for all of our users. To that end, we’re fortunate to have strong collaborative partnerships with numerous researchers, such as Ben Nassi (Confidentiality), Stav Cohen (Technion), and Or Yair (SafeBreach), as well as other AI Security researchers participating in our BugSWAT events and AI VRP program. We appreciate the work of these researchers and others in the community to help us red team and refine our defenses.


We continue working to make upcoming Gemini models inherently more resilient and add additional prompt injection defenses directly into Gemini later this year. To learn more about Google’s progress and research on generative AI threat actors, attack techniques, and vulnerabilities, take a look at the following resources:


Mitigating prompt injection attacks with a layered defense strategy

Posted by Google GenAI Security Team

With the rapid adoption of generative AI, a new wave of threats is emerging across the industry with the aim of manipulating the AI systems themselves. One such emerging attack vector is indirect prompt injections. Unlike direct prompt injections, where an attacker directly inputs malicious commands into a prompt, indirect prompt injections involve hidden malicious instructions within external data sources. These may include emails, documents, or calendar invites that instruct AI to exfiltrate user data or execute other rogue actions. As more governments, businesses, and individuals adopt generative AI to get more done, this subtle yet potentially potent attack becomes increasingly pertinent across the industry, demanding immediate attention and robust security measures.


At Google, our teams have a longstanding precedent of investing in a defense-in-depth strategy, including robust evaluation, threat analysis, AI security best practices, AI red-teaming, adversarial training, and model hardening for generative AI tools. This approach enables safer adoption of Gemini in Google Workspace and the Gemini app (we refer to both in this blog as “Gemini” for simplicity). Below we describe our prompt injection mitigation product strategy based on extensive research, development, and deployment of improved security mitigations.


A layered security approach

Google has taken a layered security approach introducing security measures designed for each stage of the prompt lifecycle. From Gemini 2.5 model hardening, to purpose-built machine learning (ML) models detecting malicious instructions, to system-level safeguards, we are meaningfully elevating the difficulty, expense, and complexity faced by an attacker. This approach compels adversaries to resort to methods that are either more easily identified or demand greater resources. 


Our model training with adversarial data significantly enhanced our defenses against indirect prompt injection attacks in Gemini 2.5 models (technical details). This inherent model resilience is augmented with additional defenses that we built directly into Gemini, including: 


  1. Prompt injection content classifiers

  2. Security thought reinforcement

  3. Markdown sanitization and suspicious URL redaction

  4. User confirmation framework

  5. End-user security mitigation notifications


This layered approach to our security strategy strengthens the overall security framework for Gemini – throughout the prompt lifecycle and across diverse attack techniques.


1. Prompt injection content classifiers


Through collaboration with leading AI security researchers via Google's AI Vulnerability Reward Program (VRP), we've curated one of the world’s most advanced catalogs of generative AI vulnerabilities and adversarial data. Utilizing this resource, we built and are in the process of rolling out proprietary machine learning models that can detect malicious prompts and instructions within various formats, such as emails and files, drawing from real-world examples. Consequently, when users query Workspace data with Gemini, the content classifiers filter out harmful data containing malicious instructions, helping to ensure a secure end-to-end user experience by retaining only safe content. For example, if a user receives an email in Gmail that includes malicious instructions, our content classifiers help to detect and disregard malicious instructions, then generate a safe response for the user. This is in addition to built-in defenses in Gmail that automatically block more than 99.9% of spam, phishing attempts, and malware.


A diagram of Gemini’s actions based on the detection of the malicious instructions by content classifiers.


2. Security thought reinforcement


This technique adds targeted security instructions surrounding the prompt content to remind the large language model (LLM) to perform the user-directed task and ignore any adversarial instructions that could be present in the content. With this approach, we steer the LLM to stay focused on the task and ignore harmful or malicious requests added by a threat actor to execute indirect prompt injection attacks.

A diagram of Gemini’s actions based on additional protection provided by the security thought reinforcement technique. 


3. Markdown sanitization and suspicious URL redaction 


Our markdown sanitizer identifies external image URLs and will not render them, making the “EchoLeak” 0-click image rendering exfiltration vulnerability not applicable to Gemini. From there, a key protection against prompt injection and data exfiltration attacks occurs at the URL level. With external data containing dynamic URLs, users may encounter unknown risks as these URLs may be designed for indirect prompt injections and data exfiltration attacks. Malicious instructions executed on a user's behalf may also generate harmful URLs. With Gemini, our defense system includes suspicious URL detection based on Google Safe Browsing to differentiate between safe and unsafe links, providing a secure experience by helping to prevent URL-based attacks. For example, if a document contains malicious URLs and a user is summarizing the content with Gemini, the suspicious URLs will be redacted in Gemini’s response. 


Gemini in Gmail provides a summary of an email thread. In the summary, there is an unsafe URL. That URL is redacted in the response and is replaced with the text “suspicious link removed”. 


4. User confirmation framework


Gemini also features a contextual user confirmation system. This framework enables Gemini to require user confirmation for certain actions, also known as “Human-In-The-Loop” (HITL), using these responses to bolster security and streamline the user experience. For example, potentially risky operations like deleting a calendar event may trigger an explicit user confirmation request, thereby helping to prevent undetected or immediate execution of the operation.


The Gemini app with instructions to delete all events on Saturday. Gemini responds with the events found on Google Calendar and asks the user to confirm this action.


5. End-user security mitigation notifications


A key aspect to keeping our users safe is sharing details on attacks that we’ve stopped so users can watch out for similar attacks in the future. To that end, when security issues are mitigated with our built-in defenses, end users are provided with contextual information allowing them to learn more via dedicated help center articles. For example, if Gemini summarizes a file containing malicious instructions and one of Google’s prompt injection defenses mitigates the situation, a security notification with a “Learn more” link will be displayed for the user. Users are encouraged to become more familiar with our prompt injection defenses by reading the Help Center article. 


Gemini in Docs with instructions to provide a summary of a file. Suspicious content was detected and a response was not provided. There is a yellow security notification banner for the user and a statement that Gemini’s response has been removed, with a “Learn more” link to a relevant Help Center article.

Moving forward


Our comprehensive prompt injection security strategy strengthens the overall security framework for Gemini. Beyond the techniques described above, it also involves rigorous testing through manual and automated red teams, generative AI security BugSWAT events, strong security standards like our Secure AI Framework (SAIF), and partnerships with both external researchers via the Google AI Vulnerability Reward Program (VRP) and industry peers via the Coalition for Secure AI (CoSAI). Our commitment to trust includes collaboration with the security community to responsibly disclose AI security vulnerabilities, share our latest threat intelligence on ways we see bad actors trying to leverage AI, and offering insights into our work to build stronger prompt injection defenses. 


Working closely with industry partners is crucial to building stronger protections for all of our users. To that end, we’re fortunate to have strong collaborative partnerships with numerous researchers, such as Ben Nassi (Confidentiality), Stav Cohen (Technion), and Or Yair (SafeBreach), as well as other AI Security researchers participating in our BugSWAT events and AI VRP program. We appreciate the work of these researchers and others in the community to help us red team and refine our defenses.


We continue working to make upcoming Gemini models inherently more resilient and add additional prompt injection defenses directly into Gemini later this year. To learn more about Google’s progress and research on generative AI threat actors, attack techniques, and vulnerabilities, take a look at the following resources:


Mitigating prompt injection attacks with a layered defense strategy

Posted by Google GenAI Security Team

With the rapid adoption of generative AI, a new wave of threats is emerging across the industry with the aim of manipulating the AI systems themselves. One such emerging attack vector is indirect prompt injections. Unlike direct prompt injections, where an attacker directly inputs malicious commands into a prompt, indirect prompt injections involve hidden malicious instructions within external data sources. These may include emails, documents, or calendar invites that instruct AI to exfiltrate user data or execute other rogue actions. As more governments, businesses, and individuals adopt generative AI to get more done, this subtle yet potentially potent attack becomes increasingly pertinent across the industry, demanding immediate attention and robust security measures.


At Google, our teams have a longstanding precedent of investing in a defense-in-depth strategy, including robust evaluation, threat analysis, AI security best practices, AI red-teaming, adversarial training, and model hardening for generative AI tools. This approach enables safer adoption of Gemini in Google Workspace and the Gemini app (we refer to both in this blog as “Gemini” for simplicity). Below we describe our prompt injection mitigation product strategy based on extensive research, development, and deployment of improved security mitigations.


A layered security approach

Google has taken a layered security approach introducing security measures designed for each stage of the prompt lifecycle. From Gemini 2.5 model hardening, to purpose-built machine learning (ML) models detecting malicious instructions, to system-level safeguards, we are meaningfully elevating the difficulty, expense, and complexity faced by an attacker. This approach compels adversaries to resort to methods that are either more easily identified or demand greater resources. 


Our model training with adversarial data significantly enhanced our defenses against indirect prompt injection attacks in Gemini 2.5 models (technical details). This inherent model resilience is augmented with additional defenses that we built directly into Gemini, including: 


  1. Prompt injection content classifiers

  2. Security thought reinforcement

  3. Markdown sanitization and suspicious URL redaction

  4. User confirmation framework

  5. End-user security mitigation notifications


This layered approach to our security strategy strengthens the overall security framework for Gemini – throughout the prompt lifecycle and across diverse attack techniques.


1. Prompt injection content classifiers


Through collaboration with leading AI security researchers via Google's AI Vulnerability Reward Program (VRP), we've curated one of the world’s most advanced catalogs of generative AI vulnerabilities and adversarial data. Utilizing this resource, we built and are in the process of rolling out proprietary machine learning models that can detect malicious prompts and instructions within various formats, such as emails and files, drawing from real-world examples. Consequently, when users query Workspace data with Gemini, the content classifiers filter out harmful data containing malicious instructions, helping to ensure a secure end-to-end user experience by retaining only safe content. For example, if a user receives an email in Gmail that includes malicious instructions, our content classifiers help to detect and disregard malicious instructions, then generate a safe response for the user. This is in addition to built-in defenses in Gmail that automatically block more than 99.9% of spam, phishing attempts, and malware.


A diagram of Gemini’s actions based on the detection of the malicious instructions by content classifiers.


2. Security thought reinforcement


This technique adds targeted security instructions surrounding the prompt content to remind the large language model (LLM) to perform the user-directed task and ignore any adversarial instructions that could be present in the content. With this approach, we steer the LLM to stay focused on the task and ignore harmful or malicious requests added by a threat actor to execute indirect prompt injection attacks.

A diagram of Gemini’s actions based on additional protection provided by the security thought reinforcement technique. 


3. Markdown sanitization and suspicious URL redaction 


Our markdown sanitizer identifies external image URLs and will not render them, making the “EchoLeak” 0-click image rendering exfiltration vulnerability not applicable to Gemini. From there, a key protection against prompt injection and data exfiltration attacks occurs at the URL level. With external data containing dynamic URLs, users may encounter unknown risks as these URLs may be designed for indirect prompt injections and data exfiltration attacks. Malicious instructions executed on a user's behalf may also generate harmful URLs. With Gemini, our defense system includes suspicious URL detection based on Google Safe Browsing to differentiate between safe and unsafe links, providing a secure experience by helping to prevent URL-based attacks. For example, if a document contains malicious URLs and a user is summarizing the content with Gemini, the suspicious URLs will be redacted in Gemini’s response. 


Gemini in Gmail provides a summary of an email thread. In the summary, there is an unsafe URL. That URL is redacted in the response and is replaced with the text “suspicious link removed”. 


4. User confirmation framework


Gemini also features a contextual user confirmation system. This framework enables Gemini to require user confirmation for certain actions, also known as “Human-In-The-Loop” (HITL), using these responses to bolster security and streamline the user experience. For example, potentially risky operations like deleting a calendar event may trigger an explicit user confirmation request, thereby helping to prevent undetected or immediate execution of the operation.


The Gemini app with instructions to delete all events on Saturday. Gemini responds with the events found on Google Calendar and asks the user to confirm this action.


5. End-user security mitigation notifications


A key aspect to keeping our users safe is sharing details on attacks that we’ve stopped so users can watch out for similar attacks in the future. To that end, when security issues are mitigated with our built-in defenses, end users are provided with contextual information allowing them to learn more via dedicated help center articles. For example, if Gemini summarizes a file containing malicious instructions and one of Google’s prompt injection defenses mitigates the situation, a security notification with a “Learn more” link will be displayed for the user. Users are encouraged to become more familiar with our prompt injection defenses by reading the Help Center article. 


Gemini in Docs with instructions to provide a summary of a file. Suspicious content was detected and a response was not provided. There is a yellow security notification banner for the user and a statement that Gemini’s response has been removed, with a “Learn more” link to a relevant Help Center article.

Moving forward


Our comprehensive prompt injection security strategy strengthens the overall security framework for Gemini. Beyond the techniques described above, it also involves rigorous testing through manual and automated red teams, generative AI security BugSWAT events, strong security standards like our Secure AI Framework (SAIF), and partnerships with both external researchers via the Google AI Vulnerability Reward Program (VRP) and industry peers via the Coalition for Secure AI (CoSAI). Our commitment to trust includes collaboration with the security community to responsibly disclose AI security vulnerabilities, share our latest threat intelligence on ways we see bad actors trying to leverage AI, and offering insights into our work to build stronger prompt injection defenses. 


Working closely with industry partners is crucial to building stronger protections for all of our users. To that end, we’re fortunate to have strong collaborative partnerships with numerous researchers, such as Ben Nassi (Confidentiality), Stav Cohen (Technion), and Or Yair (SafeBreach), as well as other AI Security researchers participating in our BugSWAT events and AI VRP program. We appreciate the work of these researchers and others in the community to help us red team and refine our defenses.


We continue working to make upcoming Gemini models inherently more resilient and add additional prompt injection defenses directly into Gemini later this year. To learn more about Google’s progress and research on generative AI threat actors, attack techniques, and vulnerabilities, take a look at the following resources:


Mitigating prompt injection attacks with a layered defense strategy

Posted by Google GenAI Security Team

With the rapid adoption of generative AI, a new wave of threats is emerging across the industry with the aim of manipulating the AI systems themselves. One such emerging attack vector is indirect prompt injections. Unlike direct prompt injections, where an attacker directly inputs malicious commands into a prompt, indirect prompt injections involve hidden malicious instructions within external data sources. These may include emails, documents, or calendar invites that instruct AI to exfiltrate user data or execute other rogue actions. As more governments, businesses, and individuals adopt generative AI to get more done, this subtle yet potentially potent attack becomes increasingly pertinent across the industry, demanding immediate attention and robust security measures.


At Google, our teams have a longstanding precedent of investing in a defense-in-depth strategy, including robust evaluation, threat analysis, AI security best practices, AI red-teaming, adversarial training, and model hardening for generative AI tools. This approach enables safer adoption of Gemini in Google Workspace and the Gemini app (we refer to both in this blog as “Gemini” for simplicity). Below we describe our prompt injection mitigation product strategy based on extensive research, development, and deployment of improved security mitigations.


A layered security approach

Google has taken a layered security approach introducing security measures designed for each stage of the prompt lifecycle. From Gemini 2.5 model hardening, to purpose-built machine learning (ML) models detecting malicious instructions, to system-level safeguards, we are meaningfully elevating the difficulty, expense, and complexity faced by an attacker. This approach compels adversaries to resort to methods that are either more easily identified or demand greater resources. 


Our model training with adversarial data significantly enhanced our defenses against indirect prompt injection attacks in Gemini 2.5 models (technical details). This inherent model resilience is augmented with additional defenses that we built directly into Gemini, including: 


  1. Prompt injection content classifiers

  2. Security thought reinforcement

  3. Markdown sanitization and suspicious URL redaction

  4. User confirmation framework

  5. End-user security mitigation notifications


This layered approach to our security strategy strengthens the overall security framework for Gemini – throughout the prompt lifecycle and across diverse attack techniques.


1. Prompt injection content classifiers


Through collaboration with leading AI security researchers via Google's AI Vulnerability Reward Program (VRP), we've curated one of the world’s most advanced catalogs of generative AI vulnerabilities and adversarial data. Utilizing this resource, we built and are in the process of rolling out proprietary machine learning models that can detect malicious prompts and instructions within various formats, such as emails and files, drawing from real-world examples. Consequently, when users query Workspace data with Gemini, the content classifiers filter out harmful data containing malicious instructions, helping to ensure a secure end-to-end user experience by retaining only safe content. For example, if a user receives an email in Gmail that includes malicious instructions, our content classifiers help to detect and disregard malicious instructions, then generate a safe response for the user. This is in addition to built-in defenses in Gmail that automatically block more than 99.9% of spam, phishing attempts, and malware.


A diagram of Gemini’s actions based on the detection of the malicious instructions by content classifiers.


2. Security thought reinforcement


This technique adds targeted security instructions surrounding the prompt content to remind the large language model (LLM) to perform the user-directed task and ignore any adversarial instructions that could be present in the content. With this approach, we steer the LLM to stay focused on the task and ignore harmful or malicious requests added by a threat actor to execute indirect prompt injection attacks.

A diagram of Gemini’s actions based on additional protection provided by the security thought reinforcement technique. 


3. Markdown sanitization and suspicious URL redaction 


Our markdown sanitizer identifies external image URLs and will not render them, making the “EchoLeak” 0-click image rendering exfiltration vulnerability not applicable to Gemini. From there, a key protection against prompt injection and data exfiltration attacks occurs at the URL level. With external data containing dynamic URLs, users may encounter unknown risks as these URLs may be designed for indirect prompt injections and data exfiltration attacks. Malicious instructions executed on a user's behalf may also generate harmful URLs. With Gemini, our defense system includes suspicious URL detection based on Google Safe Browsing to differentiate between safe and unsafe links, providing a secure experience by helping to prevent URL-based attacks. For example, if a document contains malicious URLs and a user is summarizing the content with Gemini, the suspicious URLs will be redacted in Gemini’s response. 


Gemini in Gmail provides a summary of an email thread. In the summary, there is an unsafe URL. That URL is redacted in the response and is replaced with the text “suspicious link removed”. 


4. User confirmation framework


Gemini also features a contextual user confirmation system. This framework enables Gemini to require user confirmation for certain actions, also known as “Human-In-The-Loop” (HITL), using these responses to bolster security and streamline the user experience. For example, potentially risky operations like deleting a calendar event may trigger an explicit user confirmation request, thereby helping to prevent undetected or immediate execution of the operation.


The Gemini app with instructions to delete all events on Saturday. Gemini responds with the events found on Google Calendar and asks the user to confirm this action.


5. End-user security mitigation notifications


A key aspect to keeping our users safe is sharing details on attacks that we’ve stopped so users can watch out for similar attacks in the future. To that end, when security issues are mitigated with our built-in defenses, end users are provided with contextual information allowing them to learn more via dedicated help center articles. For example, if Gemini summarizes a file containing malicious instructions and one of Google’s prompt injection defenses mitigates the situation, a security notification with a “Learn more” link will be displayed for the user. Users are encouraged to become more familiar with our prompt injection defenses by reading the Help Center article. 


Gemini in Docs with instructions to provide a summary of a file. Suspicious content was detected and a response was not provided. There is a yellow security notification banner for the user and a statement that Gemini’s response has been removed, with a “Learn more” link to a relevant Help Center article.

Moving forward


Our comprehensive prompt injection security strategy strengthens the overall security framework for Gemini. Beyond the techniques described above, it also involves rigorous testing through manual and automated red teams, generative AI security BugSWAT events, strong security standards like our Secure AI Framework (SAIF), and partnerships with both external researchers via the Google AI Vulnerability Reward Program (VRP) and industry peers via the Coalition for Secure AI (CoSAI). Our commitment to trust includes collaboration with the security community to responsibly disclose AI security vulnerabilities, share our latest threat intelligence on ways we see bad actors trying to leverage AI, and offering insights into our work to build stronger prompt injection defenses. 


Working closely with industry partners is crucial to building stronger protections for all of our users. To that end, we’re fortunate to have strong collaborative partnerships with numerous researchers, such as Ben Nassi (Confidentiality), Stav Cohen (Technion), and Or Yair (SafeBreach), as well as other AI Security researchers participating in our BugSWAT events and AI VRP program. We appreciate the work of these researchers and others in the community to help us red team and refine our defenses.


We continue working to make upcoming Gemini models inherently more resilient and add additional prompt injection defenses directly into Gemini later this year. To learn more about Google’s progress and research on generative AI threat actors, attack techniques, and vulnerabilities, take a look at the following resources:


Mitigating prompt injection attacks with a layered defense strategy

Posted by Google GenAI Security Team

With the rapid adoption of generative AI, a new wave of threats is emerging across the industry with the aim of manipulating the AI systems themselves. One such emerging attack vector is indirect prompt injections. Unlike direct prompt injections, where an attacker directly inputs malicious commands into a prompt, indirect prompt injections involve hidden malicious instructions within external data sources. These may include emails, documents, or calendar invites that instruct AI to exfiltrate user data or execute other rogue actions. As more governments, businesses, and individuals adopt generative AI to get more done, this subtle yet potentially potent attack becomes increasingly pertinent across the industry, demanding immediate attention and robust security measures.


At Google, our teams have a longstanding precedent of investing in a defense-in-depth strategy, including robust evaluation, threat analysis, AI security best practices, AI red-teaming, adversarial training, and model hardening for generative AI tools. This approach enables safer adoption of Gemini in Google Workspace and the Gemini app (we refer to both in this blog as “Gemini” for simplicity). Below we describe our prompt injection mitigation product strategy based on extensive research, development, and deployment of improved security mitigations.


A layered security approach

Google has taken a layered security approach introducing security measures designed for each stage of the prompt lifecycle. From Gemini 2.5 model hardening, to purpose-built machine learning (ML) models detecting malicious instructions, to system-level safeguards, we are meaningfully elevating the difficulty, expense, and complexity faced by an attacker. This approach compels adversaries to resort to methods that are either more easily identified or demand greater resources. 


Our model training with adversarial data significantly enhanced our defenses against indirect prompt injection attacks in Gemini 2.5 models (technical details). This inherent model resilience is augmented with additional defenses that we built directly into Gemini, including: 


  1. Prompt injection content classifiers

  2. Security thought reinforcement

  3. Markdown sanitization and suspicious URL redaction

  4. User confirmation framework

  5. End-user security mitigation notifications


This layered approach to our security strategy strengthens the overall security framework for Gemini – throughout the prompt lifecycle and across diverse attack techniques.


1. Prompt injection content classifiers


Through collaboration with leading AI security researchers via Google's AI Vulnerability Reward Program (VRP), we've curated one of the world’s most advanced catalogs of generative AI vulnerabilities and adversarial data. Utilizing this resource, we built and are in the process of rolling out proprietary machine learning models that can detect malicious prompts and instructions within various formats, such as emails and files, drawing from real-world examples. Consequently, when users query Workspace data with Gemini, the content classifiers filter out harmful data containing malicious instructions, helping to ensure a secure end-to-end user experience by retaining only safe content. For example, if a user receives an email in Gmail that includes malicious instructions, our content classifiers help to detect and disregard malicious instructions, then generate a safe response for the user. This is in addition to built-in defenses in Gmail that automatically block more than 99.9% of spam, phishing attempts, and malware.


A diagram of Gemini’s actions based on the detection of the malicious instructions by content classifiers.


2. Security thought reinforcement


This technique adds targeted security instructions surrounding the prompt content to remind the large language model (LLM) to perform the user-directed task and ignore any adversarial instructions that could be present in the content. With this approach, we steer the LLM to stay focused on the task and ignore harmful or malicious requests added by a threat actor to execute indirect prompt injection attacks.

A diagram of Gemini’s actions based on additional protection provided by the security thought reinforcement technique. 


3. Markdown sanitization and suspicious URL redaction 


Our markdown sanitizer identifies external image URLs and will not render them, making the “EchoLeak” 0-click image rendering exfiltration vulnerability not applicable to Gemini. From there, a key protection against prompt injection and data exfiltration attacks occurs at the URL level. With external data containing dynamic URLs, users may encounter unknown risks as these URLs may be designed for indirect prompt injections and data exfiltration attacks. Malicious instructions executed on a user's behalf may also generate harmful URLs. With Gemini, our defense system includes suspicious URL detection based on Google Safe Browsing to differentiate between safe and unsafe links, providing a secure experience by helping to prevent URL-based attacks. For example, if a document contains malicious URLs and a user is summarizing the content with Gemini, the suspicious URLs will be redacted in Gemini’s response. 


Gemini in Gmail provides a summary of an email thread. In the summary, there is an unsafe URL. That URL is redacted in the response and is replaced with the text “suspicious link removed”. 


4. User confirmation framework


Gemini also features a contextual user confirmation system. This framework enables Gemini to require user confirmation for certain actions, also known as “Human-In-The-Loop” (HITL), using these responses to bolster security and streamline the user experience. For example, potentially risky operations like deleting a calendar event may trigger an explicit user confirmation request, thereby helping to prevent undetected or immediate execution of the operation.


The Gemini app with instructions to delete all events on Saturday. Gemini responds with the events found on Google Calendar and asks the user to confirm this action.


5. End-user security mitigation notifications


A key aspect to keeping our users safe is sharing details on attacks that we’ve stopped so users can watch out for similar attacks in the future. To that end, when security issues are mitigated with our built-in defenses, end users are provided with contextual information allowing them to learn more via dedicated help center articles. For example, if Gemini summarizes a file containing malicious instructions and one of Google’s prompt injection defenses mitigates the situation, a security notification with a “Learn more” link will be displayed for the user. Users are encouraged to become more familiar with our prompt injection defenses by reading the Help Center article. 


Gemini in Docs with instructions to provide a summary of a file. Suspicious content was detected and a response was not provided. There is a yellow security notification banner for the user and a statement that Gemini’s response has been removed, with a “Learn more” link to a relevant Help Center article.

Moving forward


Our comprehensive prompt injection security strategy strengthens the overall security framework for Gemini. Beyond the techniques described above, it also involves rigorous testing through manual and automated red teams, generative AI security BugSWAT events, strong security standards like our Secure AI Framework (SAIF), and partnerships with both external researchers via the Google AI Vulnerability Reward Program (VRP) and industry peers via the Coalition for Secure AI (CoSAI). Our commitment to trust includes collaboration with the security community to responsibly disclose AI security vulnerabilities, share our latest threat intelligence on ways we see bad actors trying to leverage AI, and offering insights into our work to build stronger prompt injection defenses. 


Working closely with industry partners is crucial to building stronger protections for all of our users. To that end, we’re fortunate to have strong collaborative partnerships with numerous researchers, such as Ben Nassi (Confidentiality), Stav Cohen (Technion), and Or Yair (SafeBreach), as well as other AI Security researchers participating in our BugSWAT events and AI VRP program. We appreciate the work of these researchers and others in the community to help us red team and refine our defenses.


We continue working to make upcoming Gemini models inherently more resilient and add additional prompt injection defenses directly into Gemini later this year. To learn more about Google’s progress and research on generative AI threat actors, attack techniques, and vulnerabilities, take a look at the following resources:


Mitigating prompt injection attacks with a layered defense strategy

Posted by Google GenAI Security Team

With the rapid adoption of generative AI, a new wave of threats is emerging across the industry with the aim of manipulating the AI systems themselves. One such emerging attack vector is indirect prompt injections. Unlike direct prompt injections, where an attacker directly inputs malicious commands into a prompt, indirect prompt injections involve hidden malicious instructions within external data sources. These may include emails, documents, or calendar invites that instruct AI to exfiltrate user data or execute other rogue actions. As more governments, businesses, and individuals adopt generative AI to get more done, this subtle yet potentially potent attack becomes increasingly pertinent across the industry, demanding immediate attention and robust security measures.


At Google, our teams have a longstanding precedent of investing in a defense-in-depth strategy, including robust evaluation, threat analysis, AI security best practices, AI red-teaming, adversarial training, and model hardening for generative AI tools. This approach enables safer adoption of Gemini in Google Workspace and the Gemini app (we refer to both in this blog as “Gemini” for simplicity). Below we describe our prompt injection mitigation product strategy based on extensive research, development, and deployment of improved security mitigations.


A layered security approach

Google has taken a layered security approach introducing security measures designed for each stage of the prompt lifecycle. From Gemini 2.5 model hardening, to purpose-built machine learning (ML) models detecting malicious instructions, to system-level safeguards, we are meaningfully elevating the difficulty, expense, and complexity faced by an attacker. This approach compels adversaries to resort to methods that are either more easily identified or demand greater resources. 


Our model training with adversarial data significantly enhanced our defenses against indirect prompt injection attacks in Gemini 2.5 models (technical details). This inherent model resilience is augmented with additional defenses that we built directly into Gemini, including: 


  1. Prompt injection content classifiers

  2. Security thought reinforcement

  3. Markdown sanitization and suspicious URL redaction

  4. User confirmation framework

  5. End-user security mitigation notifications


This layered approach to our security strategy strengthens the overall security framework for Gemini – throughout the prompt lifecycle and across diverse attack techniques.


1. Prompt injection content classifiers


Through collaboration with leading AI security researchers via Google's AI Vulnerability Reward Program (VRP), we've curated one of the world’s most advanced catalogs of generative AI vulnerabilities and adversarial data. Utilizing this resource, we built and are in the process of rolling out proprietary machine learning models that can detect malicious prompts and instructions within various formats, such as emails and files, drawing from real-world examples. Consequently, when users query Workspace data with Gemini, the content classifiers filter out harmful data containing malicious instructions, helping to ensure a secure end-to-end user experience by retaining only safe content. For example, if a user receives an email in Gmail that includes malicious instructions, our content classifiers help to detect and disregard malicious instructions, then generate a safe response for the user. This is in addition to built-in defenses in Gmail that automatically block more than 99.9% of spam, phishing attempts, and malware.


A diagram of Gemini’s actions based on the detection of the malicious instructions by content classifiers.


2. Security thought reinforcement


This technique adds targeted security instructions surrounding the prompt content to remind the large language model (LLM) to perform the user-directed task and ignore any adversarial instructions that could be present in the content. With this approach, we steer the LLM to stay focused on the task and ignore harmful or malicious requests added by a threat actor to execute indirect prompt injection attacks.

A diagram of Gemini’s actions based on additional protection provided by the security thought reinforcement technique. 


3. Markdown sanitization and suspicious URL redaction 


Our markdown sanitizer identifies external image URLs and will not render them, making the “EchoLeak” 0-click image rendering exfiltration vulnerability not applicable to Gemini. From there, a key protection against prompt injection and data exfiltration attacks occurs at the URL level. With external data containing dynamic URLs, users may encounter unknown risks as these URLs may be designed for indirect prompt injections and data exfiltration attacks. Malicious instructions executed on a user's behalf may also generate harmful URLs. With Gemini, our defense system includes suspicious URL detection based on Google Safe Browsing to differentiate between safe and unsafe links, providing a secure experience by helping to prevent URL-based attacks. For example, if a document contains malicious URLs and a user is summarizing the content with Gemini, the suspicious URLs will be redacted in Gemini’s response. 


Gemini in Gmail provides a summary of an email thread. In the summary, there is an unsafe URL. That URL is redacted in the response and is replaced with the text “suspicious link removed”. 


4. User confirmation framework


Gemini also features a contextual user confirmation system. This framework enables Gemini to require user confirmation for certain actions, also known as “Human-In-The-Loop” (HITL), using these responses to bolster security and streamline the user experience. For example, potentially risky operations like deleting a calendar event may trigger an explicit user confirmation request, thereby helping to prevent undetected or immediate execution of the operation.


The Gemini app with instructions to delete all events on Saturday. Gemini responds with the events found on Google Calendar and asks the user to confirm this action.


5. End-user security mitigation notifications


A key aspect to keeping our users safe is sharing details on attacks that we’ve stopped so users can watch out for similar attacks in the future. To that end, when security issues are mitigated with our built-in defenses, end users are provided with contextual information allowing them to learn more via dedicated help center articles. For example, if Gemini summarizes a file containing malicious instructions and one of Google’s prompt injection defenses mitigates the situation, a security notification with a “Learn more” link will be displayed for the user. Users are encouraged to become more familiar with our prompt injection defenses by reading the Help Center article. 


Gemini in Docs with instructions to provide a summary of a file. Suspicious content was detected and a response was not provided. There is a yellow security notification banner for the user and a statement that Gemini’s response has been removed, with a “Learn more” link to a relevant Help Center article.

Moving forward


Our comprehensive prompt injection security strategy strengthens the overall security framework for Gemini. Beyond the techniques described above, it also involves rigorous testing through manual and automated red teams, generative AI security BugSWAT events, strong security standards like our Secure AI Framework (SAIF), and partnerships with both external researchers via the Google AI Vulnerability Reward Program (VRP) and industry peers via the Coalition for Secure AI (CoSAI). Our commitment to trust includes collaboration with the security community to responsibly disclose AI security vulnerabilities, share our latest threat intelligence on ways we see bad actors trying to leverage AI, and offering insights into our work to build stronger prompt injection defenses. 


Working closely with industry partners is crucial to building stronger protections for all of our users. To that end, we’re fortunate to have strong collaborative partnerships with numerous researchers, such as Ben Nassi (Confidentiality), Stav Cohen (Technion), and Or Yair (SafeBreach), as well as other AI Security researchers participating in our BugSWAT events and AI VRP program. We appreciate the work of these researchers and others in the community to help us red team and refine our defenses.


We continue working to make upcoming Gemini models inherently more resilient and add additional prompt injection defenses directly into Gemini later this year. To learn more about Google’s progress and research on generative AI threat actors, attack techniques, and vulnerabilities, take a look at the following resources:


Tracking the Cost of Quantum Factoring

Posted by Craig Gidney, Quantum Research Scientist, and Sophie Schmieg, Senior Staff Cryptography Engineer 

Google Quantum AI's mission is to build best in class quantum computing for otherwise unsolvable problems. For decades the quantum and security communities have also known that large-scale quantum computers will at some point in the future likely be able to break many of today’s secure public key cryptography algorithms, such as Rivest–Shamir–Adleman (RSA). Google has long worked with the U.S. National Institute of Standards and Technology (NIST) and others in government, industry, and academia to develop and transition to post-quantum cryptography (PQC), which is expected to be resistant to quantum computing attacks. As quantum computing technology continues to advance, ongoing multi-stakeholder collaboration and action on PQC is critical.


In order to plan for the transition from today’s cryptosystems to an era of PQC, it's important the size and performance of a future quantum computer that could likely break current cryptography algorithms is carefully characterized. Yesterday, we published a preprint demonstrating that 2048-bit RSA encryption could theoretically be broken by a quantum computer with 1 million noisy qubits running for one week. This is a 20-fold decrease in the number of qubits from our previous estimate, published in 2019. Notably, quantum computers with relevant error rates currently have on the order of only 100 to 1000 qubits, and the National Institute of Standards and Technology (NIST) recently released standard PQC algorithms that are expected to be resistant to future large-scale quantum computers. However, this new result does underscore the importance of migrating to these standards in line with NIST recommended timelines. 


Estimated resources for factoring have been steadily decreasing

Quantum computers break RSA by factoring numbers, using Shor’s algorithm. Since Peter Shor published this algorithm in 1994, the estimated number of qubits needed to run it has steadily decreased. For example, in 2012, it was estimated that a 2048-bit RSA key could be broken by a quantum computer with a billion physical qubits. In 2019, using the same physical assumptions – which consider qubits with a slightly lower error rate than Google Quantum AI’s current quantum computers – the estimate was lowered to 20 million physical qubits.



Historical estimates of the number of physical qubits needed to factor 2048-bit RSA integers.


This result represents a 20-fold decrease compared to our estimate from 2019

The reduction in physical qubit count comes from two sources: better algorithms and better error correction – whereby qubits used by the algorithm ("logical qubits") are redundantly encoded across many physical qubits, so that errors can be detected and corrected.


On the algorithmic side, the key change is to compute an approximate modular exponentiation rather than an exact one. An algorithm for doing this, while using only small work registers, was discovered in 2024 by Chevignard and Fouque and Schrottenloher. Their algorithm used 1000x more operations than prior work, but we found ways to reduce that overhead down to 2x.


On the error correction side, the key change is tripling the storage density of idle logical qubits by adding a second layer of error correction. Normally more error correction layers means more overhead, but a good combination was discovered by the Google Quantum AI team in 2023. Another notable error correction improvement is using "magic state cultivation", proposed by the Google Quantum AI team in 2024, to reduce the workspace required for certain basic quantum operations. These error correction improvements aren't specific to factoring and also reduce the required resources for other quantum computations like in chemistry and materials simulation.


Security implications

NIST recently concluded a PQC competition that resulted in the first set of PQC standards. These algorithms can already be deployed to defend against quantum computers well before a working cryptographically relevant quantum computer is built. 


To assess the security implications of quantum computers, however, it’s instructive to additionally take a closer look at the affected algorithms (see here for a detailed look): RSA and Elliptic Curve Diffie-Hellman. As asymmetric algorithms, they are used for encryption in transit, including encryption for messaging services, as well as digital signatures (widely used to prove the authenticity of documents or software, e.g. the identity of websites). For asymmetric encryption, in particular encryption in transit, the motivation to migrate to PQC is made more urgent due to the fact that an adversary can collect ciphertexts, and later decrypt them once a quantum computer is available, known as a “store now, decrypt later” attack. Google has therefore been encrypting traffic both in Chrome and internally, switching to the standardized version of ML-KEM once it became available. Notably not affected is symmetric cryptography, which is primarily deployed in encryption at rest, and to enable some stateless services.


For signatures, things are more complex. Some signature use cases are similarly urgent, e.g., when public keys are fixed in hardware. In general, the landscape for signatures is mostly remarkable due to the higher complexity of the transition, since signature keys are used in many different places, and since these keys tend to be longer lived than the usually ephemeral encryption keys. Signature keys are therefore harder to replace and much more attractive targets to attack, especially when compute time on a quantum computer is a limited resource. This complexity likewise motivates moving earlier rather than later. To enable this, we have added PQC signature schemes in public preview in Cloud KMS. 


The initial public draft of the NIST internal report on the transition to post-quantum cryptography standards states that vulnerable systems should be deprecated after 2030 and disallowed after 2035. Our work highlights the importance of adhering to this recommended timeline.



More from Google on PQC: https://cloud.google.com/security/resources/post-quantum-cryptography?e=48754805 


Google announces Sec-Gemini v1, a new experimental cybersecurity model

Posted by Elie Burzstein and Marianna Tishchenko, Sec-Gemini team



Today, we’re announcing Sec-Gemini v1, a new experimental AI model focused on advancing cybersecurity AI frontiers. 



As outlined a year ago, defenders face the daunting task of securing against all cyber threats, while attackers need to successfully find and exploit only a single vulnerability. This fundamental asymmetry has made securing systems extremely difficult, time consuming and error prone. AI-powered cybersecurity workflows have the potential to help shift the balance back to the defenders by force multiplying cybersecurity professionals like never before.


 

Effectively powering SecOps workflows requires state-of-the-art reasoning capabilities and extensive current cybersecurity knowledge. Sec-Gemini v1 achieves this by combining Gemini’s advanced capabilities with near real-time cybersecurity knowledge and tooling. This combination allows it to achieve superior performance on key cybersecurity workflows, including incident root cause analysis, threat analysis, and vulnerability impact understanding.



We firmly believe that successfully pushing AI cybersecurity frontiers to decisively tilt the balance in favor of the defenders requires a strong collaboration across the cybersecurity community. This is why we are making Sec-Gemini v1 freely available to select organizations, institutions, professionals, and NGOs for research purposes.



Sec-Gemini v1 outperforms other models on key cybersecurity benchmarks as a result of its advanced integration of Google Threat Intelligence (GTI), OSV, and other key data sources. Sec-Gemini v1 outperforms other models on CTI-MCQ, a leading threat intelligence benchmark, by at least 11% (See Figure 1). It also outperforms other models by at least 10.5% on the CTI-Root Cause Mapping benchmark (See Figure 2):





Figure 1: Sec-Gemini v1 outperforms other models on the CTI-MCQ Cybersecurity Threat Intelligence benchmark.







Figure 2: Sec-Gemini v1 has outperformed other models in a Cybersecurity Threat Intelligence-Root Cause Mapping (CTI-RCM) benchmark that evaluates an LLM's ability to understand the nuances of vulnerability descriptions, identify vulnerabilities underlying root causes, and accurately classify them according to the CWE taxonomy.




Below is an example of the comprehensiveness of Sec-Gemini v1’s answers in response to key cybersecurity questions. First, Sec-Gemini v1 is able to determine that Salt Typhoon is a threat actor (not all models do) and provides a comprehensive description of that threat actor, thanks to its deep integration with Mandiant Threat intelligence data.









Next, in response to a question about the vulnerabilities in the Salt Typhoon description, Sec-Gemini v1 outputs not only vulnerability details (thanks to its integration with OSV data, the open-source vulnerabilities database operated by Google), but also contextualizes the vulnerabilities with respect to threat actors (using Mandiant data). With Sec-Gemini v1, analysts can understand the risk and threat profile associated with specific vulnerabilities faster.








If you are interested in collaborating with us on advancing the AI cybersecurity frontier, please request early access to Sec-Gemini v1 via this form.







Taming the Wild West of ML: Practical Model Signing with Sigstore

Posted by Mihai Maruseac, Google Open Source Security Team (GOSST)


In partnership with NVIDIA and HiddenLayer, as part of the Open Source Security Foundation, we are now launching the first stable version of our model signing library. Using digital signatures like those from Sigstore, we allow users to verify that the model used by the application is exactly the model that was created by the developers. In this blog post we will illustrate why this release is important from Google’s point of view.



With the advent of LLMs, the ML field has entered an era of rapid evolution. We have seen remarkable progress leading to weekly launches of various applications which incorporate ML models to perform tasks ranging from customer support, software development, and even performing security critical tasks.



However, this has also opened the door to a new wave of security threats. Model and data poisoning, prompt injection, prompt leaking and prompt evasion are just a few of the risks that have recently been in the news. Garnering less attention are the risks around the ML supply chain process: since models are an uninspectable collection of weights (sometimes also with arbitrary code), an attacker can tamper with them and achieve significant impact to those using the models. Users, developers, and practitioners need to examine an important question during their risk assessment process: “can I trust this model?”



Since its launch, Google’s Secure AI Framework (SAIF) has created guidance and technical solutions for creating AI applications that users can trust. A first step in achieving trust in the model is to permit users to verify its integrity and provenance, to prevent tampering across all processes from training to usage, via cryptographic signing. 



The ML supply chain

To understand the need for the model signing project, let’s look at the way ML powered applications are developed, with an eye to where malicious tampering can occur.



Applications that use advanced AI models are typically developed in at least three different stages. First, a large foundation model is trained on large datasets. Next, a separate ML team finetunes the model to make it achieve good performance on application specific tasks. Finally,  this fine-tuned model is embedded into an application.



The three steps involved in building an application that uses large language models.



These three stages are usually handled by different teams, and potentially even different companies, since each stage requires specialized expertise. To make models available from one stage to the next, practitioners leverage model hubs, which are repositories for storing models. Kaggle and HuggingFace are popular open source options, although internal model hubs could also be used.



This separation into stages creates multiple opportunities where a malicious user (or external threat actor who has compromised the internal infrastructure) could tamper with the model. This could range from just a slight alteration of the model weights that control model behavior, to injecting architectural backdoors — completely new model behaviors and capabilities that could be triggered only on specific inputs. It is also possible to exploit the serialization format and inject arbitrary code execution in the model as saved on disk — our whitepaper on AI supply chain integrity goes into more details on how popular model serialization libraries could be exploited. The following diagram summarizes the risks across the ML supply chain for developing a single model, as discussed in the whitepaper.



The supply chain diagram for building a single model, illustrating some supply chain risks (oval labels) and where model signing can defend against them (check marks)



The diagram shows several places where the model could be compromised. Most of these could be prevented by signing the model during training and verifying integrity before any usage, in every step: the signature would have to be verified when the model gets uploaded to a model hub, when the model gets selected to be deployed into an application (embedded or via remote APIs) and when the model is used as an intermediary during another training run. Assuming the training infrastructure is trustworthy and not compromised, this approach guarantees that each model user can trust the model.



Sigstore for ML models

Signing models is inspired by code signing, a critical step in traditional software development. A signed binary artifact helps users identify its producer and prevents tampering after publication. The average developer, however, would not want to manage keys and rotate them on compromise.



These challenges are addressed by using Sigstore, a collection of tools and services that make code signing secure and easy. By binding an OpenID Connect token to a workload or developer identity, Sigstore alleviates the need to manage or rotate long-lived secrets. Furthermore, signing is made transparent so signatures over malicious artifacts could be audited in a public transparency log, by anyone. This ensures that split-view attacks are not possible, so any user would get the exact same model. These features are why we recommend Sigstore’s signing mechanism as the default approach for signing ML models.



Today the OSS community is releasing the v1.0 stable version of our model signing library as a Python package supporting Sigstore and traditional signing methods. This model signing library is specialized to handle the sheer scale of ML models (which are usually much larger than traditional software components), and handles signing models represented as a directory tree. The package provides CLI utilities so that users can sign and verify model signatures for individual models. The package can also be used as a library which we plan to incorporate directly into model hub upload flows as well as into ML frameworks.



Future goals

We can view model signing as establishing the foundation of trust in the ML ecosystem. We envision extending this approach to also include datasets and other ML-related artifacts. Then, we plan to build on top of signatures, towards fully tamper-proof metadata records, that can be read by both humans and machines. This has the potential to automate a significant fraction of the work needed to perform incident response in case of a compromise in the ML world. In an ideal world, an ML developer would not need to perform any code changes to the training code, while the framework itself would handle model signing and verification in a transparent manner.



If you are interested in the future of this project, join the OpenSSF meetings attached to the project. To shape the future of building tamper-proof ML, join the Coalition for Secure AI, where we are planning to work on building the entire trust ecosystem together with the open source community. In collaboration with multiple industry partners, we are starting up a special interest group under CoSAI for defining the future of ML signing and including tamper-proof ML metadata, such as model cards and evaluation results.

Titan Security Keys now available in more countries

Posted by Christiaan Brand, Group Product Manager

We’re excited to announce that starting today, Titan Security Keys are available for purchase in more than 10 new countries:

  • Ireland

  • Portugal

  • The Netherlands

  • Denmark

  • Norway

  • Sweden

  • Finland

  • Australia

  • New Zealand

  • Singapore

  • Puerto Rico

This expansion means Titan Security Keys are now available in 22 markets, including previously announced countries like Austria, Belgium, Canada, France, Germany, Italy, Japan, Spain, Switzerland, the UK, and the US.


What is a Titan Security Key?

A Titan Security Key is a small, physical device that you can use to verify your identity when you sign in to your Google Account. It’s like a second password that’s much harder for cybercriminals to steal.

Titan Security Keys allow you to store your passkeys on a strong, purpose-built device that can help protect you against phishing and other online attacks. They’re easy to use and work with a wide range of devices and services as they’re compatible with the FIDO2 standard.

How do I use a Titan Security Key?

To use a Titan Security Key, you simply plug it into your computer’s USB port or tap it to your device using NFC. When you’re asked to verify your identity, you’ll just need to tap the button on the key.

Where can I buy a Titan Security Key?

You can buy Titan Security Keys on the Google Store.


We’re committed to making our products available to as many people as possible and we hope this expansion will help more people stay safe online.