Humans Behind Search: Meet Matt

Matt Cooke, who heads up the Google News Lab, talks about how his team’s keeping it real in news and search.

First, can you tell us about the Google News Initiative and the work that your team does?

The Google News Initiative is the part of Google that works with journalists and entrepreneurs to drive innovation in news — from surfacing factual information to helping local publishers to digitize their content.

As part of that, the News Lab offers partnerships and training in 70 countries around the world to bring Google technology to journalists and news publishers. We want to help strengthen digital skills to help journalists verify sources, fact check and explore different forms of storytelling for audiences searching for accurate information.

Tell us about your background and what led to working at Google.

I worked for a number of years at BBC in various roles, including reporting from the East London multimedia newsroom in the build-up to the 2012 Olympic Games. We gave members of the audience access to small cameras, which got me thinking about the potential of digital storytelling. So when the opportunity came along at Google, I seized it! I’m coming up to my 10th anniversary here.

What tips can you give readers searching for credible news and information?

News comes to us in so many different ways and formats these days so it can sometimes be hard to tell what's authoritative and accurate. But there are tools available to help, and there are five things I would recommend when it comes to checking authenticity:

  1. If you’ve stumbled across something surprising, check the source. The About This Result feature provides details about a website before you visit it, including its description, when it was first indexed and whether your connection to the site is secure.
  2. If an image looks suspicious you can go toGoogle Images and do a reverse image search by clicking on the camera icon and dragging in your picture. An example of an obvious fake, that many will have seen and that recurs, is the shark swimming up a flooded street.
  3. Again if a story is surprising, look at other news sources to see if they’re covering it too. If they’re not, that could give you some pause for thought. Stories on Google news have the option for ‘full coverage,’ which means you can see how others are reporting the same story.
  4. We have something called theFact Check Explorer which allows you to type in a search term and then it shows you counterclaims and debunks of that theory by fact checking organisations.
  5. Google Maps, Google Earth and Street View can all help you verify whether an image that you are seeing is from the reported location. This can be done by checking for shadows of, for example, mountains or nearby buildings on Street View, and by matching them up with Google Earth.

There are some amazing and painstaking examples of work from BBC Africa Eye with video verification. It shows that by combining digital tools with great journalism, you can get to the heart of what’s really going on.

How are you tackling deep fakes?

Technology evolves fast, but our engineering teams and trust and safety teams are working hard to make sure that our technology stays ahead of emerging threats, and we’re working closely with industry partners too to understand how they’re engaging with verification and misinformation. It’s an industry-wide effort.

Progress is being made, for example YouTube has something calledContent ID which allows organisations to detect their IP copyright and content.

How does Google make sure that organisations don’t game the algorithm?

We update the search algorithm thousands of times a year and work hard to make sure that the information people find is the most accurate, authoritative and relevant to what they’re looking for. We have policies in place to prevent spam and any deceptive practices — there’s more information here about how Search works. And I’d also add that we place a lot of importance on people finding good quality local news from reliable sources as well.

How do you make sure that search literacy isn’t confined to the few and privileged?

Over the last few years we’ve spent a lot of time with media literacy experts to provide newsroom style lessons for primary and secondary school students across the UK, a good example is Be Internet Legends which has reached over 70% of primary schools in England.

And there’s more that can be done outside the classroom. Last year, we contributed €25 million to the European Media and Information Fund, which supports funding for publishers, academics and researchers to research and implement media literacy.

What excites you about Search?

One of the things that my team works with is a tool calledGoogle Trends, which shows you anonymised, indexed data on what people are searching for across a given location or a given time.

We collaborate with broadcasters, journalists, academics and non-profits to see what emerging patterns and trends in search reveal. What’s really interesting is how changing questions reveal changing attitudes across time. It’s useful for journalists, or indeed anybody who’s interested, to see what preoccupations people have around major news events.

For example, if you're looking for the latest design trends, we can see that Search interest for 'japandi interior design' has increased by 973% in the last 12 months. Similarly, if you're wondering what people in the UK are reading, we can share that between 2017 and 2021, Search interest for books on neurodiversity has increased by 1280%.

What’s your most loved search feature that everybody should know about?

Some of the more Advanced Search features can really save time. So if you’re searching for information about an institution or a public figure, often you’ll have to wade through lots of information. If you want to delve deeper you can start to remove repetitive information, simply by putting a minus symbol in front of a word. For example, I travel a lot for work so if you’re looking for information about UK airports but you don’t want to focus on Heathrow you can type in ‘UK airports -Heathrow.’ It is simple, but saves lots of time!

On Google Street View, my favourite desktop tool is the little clock at the top of the screen. This allows you to go back in time so you can see how things have changed. For example, when Britain was getting ready to host the 2012 Olympic Games, you can see all the flags and banners up in central London. And where cities change fast, for example East London, you can see how new developments have taken the place of old buildings. Take a look!

Chrome Dev for Android Update

Hi everyone! We've just released Chrome Dev 103 (103.0.5038.0) for Android. It's now available on Google Play.

You can see a partial list of the changes in the Git log. For details on new features, check out the Chromium blog, and for details on web platform updates, check here.

If you find a new issue, please let us know by filing a bug.

Krishna Govind
Google Chrome

Alpa: Automated Model-Parallel Deep Learning

Over the last several years, the rapidly growing size of deep learning models has quickly exceeded the memory capacity of single accelerators. Earlier models like BERT (with a parameter size of < 1GB) can efficiently scale across accelerators by leveraging data parallelism in which model weights are duplicated across accelerators while only partitioning and distributing the training data. However, recent large models like GPT-3 (with a parameter size of 175GB) can only scale using model parallel training, where a single model is partitioned across different devices.

While model parallelism strategies make it possible to train large models, they are more complex in that they need to be specifically designed for target neural networks and compute clusters. For example, Megatron-LM uses a model parallelism strategy to split the weight matrices by rows or columns and then synchronizes results among devices. Device placement or pipeline parallelism partitions different operators in a neural network into multiple groups and the input data into micro-batches that are executed in a pipelined fashion. Model parallelism often requires significant effort from system experts to identify an optimal parallelism plan for a specific model. But doing so is too onerous for most machine learning (ML) researchers whose primary focus is to run a model and for whom the model’s performance becomes a secondary priority. As such, there remains an opportunity to automate model parallelism so that it can easily be applied to large models.

In “Alpa: Automating Inter- and Intra-Operator Parallelism for Distributed Deep Learning”, published at OSDI 2022, we describe a method for automating the complex model parallelism process. We demonstrate that with only one line of code Alpa can transform any JAX neural network into a distributed version with an optimal parallelization strategy that can be executed on a user-provided device cluster. We are also excited to release Alpa’s code to the broader research community.

Alpa Design
We begin by grouping existing ML parallelization strategies into two categories, inter-operator parallelism and intra-operator parallelism. Inter-operator parallelism assigns distinct operators to different devices (e.g., device placement) that are often accelerated with a pipeline execution schedule (e.g., pipeline parallelism). With intra-operator parallelism, which includes data parallelism (e.g., Deepspeed-Zero), operator parallelism (e.g., Megatron-LM), and expert parallelism (e.g., GShard-MoE), individual operators are split and executed on multiple devices, and often collective communication is used to synchronize the results across devices.

The difference between these two approaches maps naturally to the heterogeneity of a typical compute cluster. Inter-operator parallelism has lower communication bandwidth requirements because it is only transmitting activations between operators on different accelerators. But, it suffers from device underutilization because of its pipeline data dependency, i.e., some operators are inactive while waiting on the outputs from other operators. In contrast, intra-operator parallelism doesn’t have the data dependency issue, but requires heavier communication across devices. In a GPU cluster, the GPUs within a node have higher communication bandwidth that can accommodate intra-operator parallelism. However, GPUs across different nodes are often connected with much lower bandwidth (e.g., ethernet) so inter-operator parallelism is preferred.

By leveraging heterogeneous mapping, we design Alpa as a compiler that conducts various passes when given a computational graph and a device cluster from a user. First, the inter-operator pass slices the computational graph into subgraphs and the device cluster into submeshes (i.e., a partitioned device cluster) and identifies the best way to assign a subgraph to a submesh. Then, the intra-operator pass finds the best intra-operator parallelism plan for each pipeline stage from the inter-operator pass. Finally, the runtime orchestration pass generates a static plan that orders the computation and communication and executes the distributed computational graph on the actual device cluster.

An overview of Alpa. In the sliced subgraphs, red and blue represent the way the operators are partitioned and gray represents operators that are replicated. Green represents the actual devices (e.g., GPUs).

Intra-Operator Pass
Similar to previous research (e.g., Mesh-TensorFlow and GSPMD), intra-operator parallelism partitions a tensor on a device mesh. This is shown below for a typical 3D tensor in a Transformer model with a given batch, sequence, and hidden dimensions. The batch dimension is partitioned along device mesh dimension 0 (mesh0), the hidden dimension is partitioned along mesh dimension 1 (mesh1), and the sequence dimension is replicated to each processor.

A 3D tensor that is partitioned on a 2D device mesh.

With the partitions of tensors in Alpa, we further define a set of parallelization strategies for each individual operator in a computational graph. We show example parallelization strategies for matrix multiplication in the figure below. Defining parallelization strategies on operators leads to possible conflicts on the partitions of tensors because one tensor can be both the output of one operator and the input of another. In this case, re-partition is needed between the two operators, which incurs additional communication costs.

The parallelization strategies for matrix multiplication.

Given the partitions of each operator and re-partition costs, we formulate the intra-operator pass as a Integer-Linear Programming (ILP) problem. For each operator, we define a one-hot variable vector to enumerate the partition strategies. The ILP objective is to minimize the sum of compute and communication cost (node cost) and re-partition communication cost (edge cost). The solution of the ILP translates to one specific way to partition the original computational graph.

Inter-Operator Pass
The inter-operator pass slices the computational graph and device cluster for pipeline parallelism. As shown below, the boxes represent micro-batches of input and the pipeline stages represent a submesh executing a subgraph. The horizontal dimension represents time and shows the pipeline stage at which a micro-batch is executed. The goal of the inter-operator pass is to minimize the total execution latency, which is the sum of the entire workload execution on the device as illustrated in the figure below. Alpa uses a Dynamic Programming (DP) algorithm to minimize the total latency. The computational graph is first flattened, and then fed to the intra-operator pass where the performance of all possible partitions of the device cluster into submeshes are profiled.

Pipeline parallelism. For a given time, this figure shows the micro-batches (colored boxes) that a partitioned device cluster and a sliced computational graph (e.g., stage 1, 2, 3) is processing.

Runtime Orchestration
After the inter- and intra-operator parallelization strategies are complete, the runtime generates and dispatches a static sequence of execution instructions for each device submesh. These instructions include RUN a specific subgraph, SEND/RECEIVE tensors from other meshes, or DELETE a specific tensor to free the memory. The devices can execute the computational graph without other coordination by following the instructions.

Evaluation
We test Alpa with eight AWS p3.16xlarge instances, each of which has eight 16 GB V100 GPUs, for 64 total GPUs. We examine weak scaling results of growing the model size while increasing the number of GPUs. We evaluate three models: (1) the standard Transformer model (GPT); (2) the GShard-MoE model, a transformer with mixture-of-expert layers; and (3) Wide-ResNet, a significantly different model with no existing expert-designed model parallelization strategy. The performance is measured by peta-floating point operations per second (PFLOPS) achieved on the cluster.

We demonstrate that for GPT, Alpa outputs a parallelization strategy very similar to the one computed by the best existing framework, Megatron-ML, and matches its performance. For GShard-MoE, Alpa outperforms the best expert-designed baseline on GPU (i.e., Deepspeed) by up to 8x. Results for Wide-ResNet show that Alpa can generate the optimal parallelization strategy for models that have not been studied by experts. We also show the linear scaling numbers for reference.

GPT: Alpa matches the performance of Megatron-ML, the best expert-designed framework.
GShard MoE: Alpa outperforms Deepspeed (the best expert-designed framework on GPU) by up to 8x.
Wide-ResNet: Alpa generalizes to models without manual plans. Pipeline and Data Parallelism (PP-DP) is a baseline model that uses only pipeline and data parallelism but no other intra-operator parallelism.
The parallelization strategy for Wide-ResNet on 16 GPUs consists of three pipeline stages and is a complicated strategy even for an expert to design. Stages 1 and 2 are on 4 GPUs performing data parallelism, and stage 3 is on 8 GPUs performing operator parallelism.

Conclusion
The process of designing an effective parallelization plan for distributed model-parallel deep learning has historically been a difficult and labor-intensive task. Alpa is a new framework that leverages intra- and inter-operator parallelism for automated model-parallel distributed training. We believe that Alpa will democratize distributed model-parallel learning and accelerate the development of large deep learning models. Explore the open-source code and learn more about Alpa in our paper.

Acknowledgements
Thanks to the co-authors of the paper: Lianmin Zheng, Hao Zhang, Yonghao Zhuang, Yida Wang, Danyang Zhuo, Joseph E. Gonzalez, and Ion Stoica. We would also like to thank Shibo Wang, Jinliang Wei, Yanping Huang, Yuanzhong Xu, Zhifeng Chen, Claire Cui, Naveen Kumar, Yash Katariya, Laurent El Shafey, Qiao Zhang, Yonghui Wu, Marcello Maggioni, Mingyao Yang, Michael Isard, Skye Wanderman-Milne, and David Majnemer for their collaborations to this research.



Source: Google AI Blog


Share your video feed when using Companion mode in Google Meet

Quick summary 

When using Companion mode in Google Meet, you can now turn your camera on and share your video feed with all other participants. For in-room participants attending a hybrid meeting, this feature helps improve collaboration and representation equity by giving everyone the ability to share their own video with other on-call participants.


Getting started

  • Admins: There is no admin control for this feature.
  • End users: 
    • This feature will be available by default. You can join a meeting on the web using Companion mode from the green room before your meeting. To share your video feed, select “Turn on camera” from the Meet toolbar.
    • Use this Help Center article and video guide to learn more about using Companion mode in Google Meet.

Rollout pace

Share your video feed when using Companion mode in Google Meet

Quick summary 

When using Companion mode in Google Meet, you can now turn your camera on and share your video feed with all other participants. For in-room participants attending a hybrid meeting, this feature helps improve collaboration and representation equity by giving everyone the ability to share their own video with other on-call participants.


Getting started

  • Admins: There is no admin control for this feature.
  • End users: 
    • This feature will be available by default. You can join a meeting on the web using Companion mode from the green room before your meeting. To share your video feed, select “Turn on camera” from the Meet toolbar.
    • Use this Help Center article and video guide to learn more about using Companion mode in Google Meet.

Rollout pace

Google Ads Scripts, AdWords API and Google Ads API reporting issues on April 25 and 26, 2022

Between April 25th 2:32 PM PT and April 26th 12:24 PM PT, there was an issue which may have impacted some read report requests across Google Ads scripts, the AdWords API, and the Google Ads API. If you were using these products to request reporting data for your accounts, then a small percentage of report downloads may have been missing rows or may have had incorrect data in a given row. This issue has been resolved. As a precaution, we recommend running again any reports that you have executed during this period as the missing data has been restored.

If you have any questions, please contact us via the Google Ads API forum or the Google Ads scripts forum.

Mothers.day: Highlighting inequality in maternal health

The path to parenthood looks different for everyone, but one element of becoming a parent is universal: the need for quality healthcare and community support. Sadly, this basic need is out of reach for far too many people. Every day, more than 800 people around the world die from pregnancy- and childbirth-related causes that could have been prevented, according to the World Health Organization.

Google Registry launched the .day top-level domain earlier this year, and today we’re introducing mothers.day — a resource dedicated to highlighting inequities in maternal health and helping families at different stages of parenthood. The website also lists ways you and your loved ones can help bridge these gaps by volunteering or donating to organizations making an impact in this space.

This year, I've asked my family to make giving to others the focus of our Mother’s Day celebration. To help pass on the value of generosity, the mothers.day website points to several nonprofits for Mother’s Day giving, including:

  • Postpartum Support International is the world’s leading nonprofit organization dedicated to helping women suffering from perinatal mood and anxiety disorders, including postpartum depression.
  • Black Mamas Matter Alliance is a Black women-led group that advocates, drives research, builds power, and shifts culture for Black maternal health, rights and justice.
  • Fistula Foundation provides life-transforming surgery to women injured in childbirth who are left incontinent and often shunned.
  • The Cradle is a nonprofit, licensed child welfare agency providing adoption services, counseling and education and a nursery for birth parents and adoptive families.
  • Hello Neighbor's Smart Start program provides refugee and immigrant mothers with socio-emotional, logistical, and material need support throughout pregnancy and postpartum.

These are just a few organizations committed to making the journey to parenthood equitable for everyone. In addition to giving, mothers.day includes information on how you can make an impact on maternal healthcare by participating in research studies:

  • Powermom is a mobile research platform with the goal of addressing health disparities and partnering with all participants during pregnancy and the postpartum period.
  • PM3 study is a study for Black women, by Black women and helps new moms in the state of Georgia stay healthy after pregnancy.
  • Maternal Near Miss aims to gather insights from women of color who've had near-death experiences during pregnancy and/or childbirth in order to inform maternal health policies and clinical practices.

There are so many ways to support birthing people and their families around the world. For more ways to get involved, visit mothers.day.

Mothers.day: Highlighting inequality in maternal health

The path to parenthood looks different for everyone, but one element of becoming a parent is universal: the need for quality healthcare and community support. Sadly, this basic need is out of reach for far too many people. Every day, more than 800 people around the world die from pregnancy- and childbirth-related causes that could have been prevented, according to the World Health Organization.

Google Registry launched the .day top-level domain earlier this year, and today we’re introducing mothers.day — a resource dedicated to highlighting inequities in maternal health and helping families at different stages of parenthood. The website also lists ways you and your loved ones can help bridge these gaps by volunteering or donating to organizations making an impact in this space.

This year, I've asked my family to make giving to others the focus of our Mother’s Day celebration. To help pass on the value of generosity, the mothers.day website points to several nonprofits for Mother’s Day giving, including:

  • Postpartum Support International is the world’s leading nonprofit organization dedicated to helping women suffering from perinatal mood and anxiety disorders, including postpartum depression.
  • Black Mamas Matter Alliance is a Black women-led group that advocates, drives research, builds power, and shifts culture for Black maternal health, rights and justice.
  • Fistula Foundation provides life-transforming surgery to women injured in childbirth who are left incontinent and often shunned.
  • The Cradle is a nonprofit, licensed child welfare agency providing adoption services, counseling and education and a nursery for birth parents and adoptive families.
  • Hello Neighbor's Smart Start program provides refugee and immigrant mothers with socio-emotional, logistical, and material need support throughout pregnancy and postpartum.

These are just a few organizations committed to making the journey to parenthood equitable for everyone. In addition to giving, mothers.day includes information on how you can make an impact on maternal healthcare by participating in research studies:

  • Powermom is a mobile research platform with the goal of addressing health disparities and partnering with all participants during pregnancy and the postpartum period.
  • PM3 study is a study for Black women, by Black women and helps new moms in the state of Georgia stay healthy after pregnancy.
  • Maternal Near Miss aims to gather insights from women of color who've had near-death experiences during pregnancy and/or childbirth in order to inform maternal health policies and clinical practices.

There are so many ways to support birthing people and their families around the world. For more ways to get involved, visit mothers.day.

Update on cyber activity in Eastern Europe

Google’s Threat Analysis Group (TAG) has been closely monitoring the cybersecurity activity in Eastern Europe with regard to the war in Ukraine. Since our last update, TAG has observed a continuously growing number of threat actors using the war as a lure in phishing and malware campaigns. Similar to other reports, we have also observed threat actors increasingly target critical infrastructure entities including oil and gas, telecommunications and manufacturing.

Government-backed actors from China, Iran, North Korea and Russia, as well as various unattributed groups, have used various Ukraine war-related themes in an effort to get targets to open malicious emails or click malicious links. Financially motivated and criminal actors are also using current events as a means for targeting users.

As always, we continue to publish details surrounding the actions we take against coordinated influence operations in our quarterly TAG bulletin. We promptly identify and remove any such content but have not observed any significant shifts from the normal levels of activity that occur in the region.

Here is a deeper look at the campaign activity TAG has observed and the actions the team has taken to protect our users over the past few weeks:

APT28 or Fancy Bear, a threat actor attributed to Russia GRU, was observed targeting users in Ukraine with a new variant of malware. The malware, distributed via email attachments inside of password protected zip files (ua_report.zip), is a .Net executable that when executed steals cookies and saved passwords from Chrome, Edge and Firefox browsers. The data is then exfiltrated via email to a compromised email account.

Malware samples:

TAG would like to thank the Yahoo! Paranoids Advanced Cyber Threats Team for their collaboration in this investigation.

Turla, a group TAG attributes to Russia FSB, continues to run campaigns against the Baltics, targeting defense and cybersecurity organizations in the region. Similar to recently observed activity, these campaigns were sent via email and contained a unique link per target that led to a DOCX file hosted on attacker controlled infrastructure. When opened, the DOCX file would attempt to download a unique PNG file from the same attacker controlled domain.

Recently observed Turla domains:

  • wkoinfo.webredirect[.]org
  • jadlactnato.webredirect[.]org

COLDRIVER, a Russian-based threat actor sometimes referred to as Callisto, continues to use Gmail accounts to send credential phishing emails to a variety of Google and non-Google accounts. The targets include government and defense officials, politicians, NGOs and think tanks, and journalists. The group's tactics, techniques and procedures (TTPs) for these campaigns have shifted slightly from including phishing links directly in the email, to also linking to PDFs and/or DOCs hosted on Google Drive and Microsoft One Drive. Within these files is a link to an attacker controlled phishing domain.

These phishing domains have been blocked through Google Safe Browsing – a service that identifies unsafe websites across the web and notifies users and website owners of potential harm.

An example of this technique

An example of this technique

Recently observed COLDRIVER credential phishing domains:

  • cache-dns[.]com
  • docs-shared[.]com
  • documents-forwarding[.]com
  • documents-preview[.]com
  • protection-link[.]online
  • webresources[.]live

Ghostwriter, a Belarusian threat actor, has remained active during the course of the war and recently resumed targeting of Gmail accounts via credential phishing. This campaign, targeting high risk individuals in Ukraine, contained links leading to compromised websites where the first stage phishing page was hosted. If the user clicked continue, they would be redirected to an attacker controlled site that collected the users credentials. There were no accounts compromised from this campaign and Google will alert all targeted users of these attempts through our monthly government-backed attacker warnings.

Both pages from this campaign are shown below.

an example webpage
An example page

In mid-April, TAG detected a Ghostwriter credential phishing campaign targeting Facebook users. The targets, primarily located in Lithuania, were sent links to attacker controlled domains from a domain spoofing the Facebook security team.

Facebook campaign

Recently observed Ghostwriter credential phishing domains and emails:

  • noreply.accountsverify[.]top
  • microsoftonline.email-verify[.]top
  • lt-microsoftgroup.serure-email[.]online
  • facebook.com-validation[.]top
  • lt-meta.com-verification[.]top
  • lt-facebook.com-verification[.]top
  • secure@facebookgroup[.]lt

Curious Gorge, a group TAG attributes to China's PLA SSF, has remained active against government, military, logistics and manufacturing organizations in Ukraine, Russia and Central Asia. In Russia, long running campaigns against multiple government organizations have continued, including the Ministry of Foreign Affairs. Over the past week, TAG identified additional compromises impacting multiple Russian defense contractors and manufacturers and a Russian logistics company.

Protecting Our Users

Upon discovery, all identified websites and domains were added to Safe Browsing to protect users from further exploitation. We also send all targeted Gmail and Workspace users government-backed attacker alerts notifying them of the activity. We encourage any potential targets to enable Google Account Level Enhanced Safe Browsing and ensure that all devices are updated.

The team continues to work around the clock, focusing on the safety and security of our users and the platforms that help them access and share important information. We’ll continue to take action, identify bad actors and share relevant information with others across industry and governments, with the goal of bringing awareness to these issues, protecting users and preventing future attacks. While we are actively monitoring activity related to Ukraine and Russia, we continue to be just as vigilant in relation to other threat actors globally, to ensure that they do not take advantage of everyone’s focus on this region.

Update on cyber activity in Eastern Europe

Google’s Threat Analysis Group (TAG) has been closely monitoring the cybersecurity activity in Eastern Europe with regard to the war in Ukraine. Since our last update, TAG has observed a continuously growing number of threat actors using the war as a lure in phishing and malware campaigns. Similar to other reports, we have also observed threat actors increasingly target critical infrastructure entities including oil and gas, telecommunications and manufacturing.

Government-backed actors from China, Iran, North Korea and Russia, as well as various unattributed groups, have used various Ukraine war-related themes in an effort to get targets to open malicious emails or click malicious links. Financially motivated and criminal actors are also using current events as a means for targeting users.

As always, we continue to publish details surrounding the actions we take against coordinated influence operations in our quarterly TAG bulletin. We promptly identify and remove any such content but have not observed any significant shifts from the normal levels of activity that occur in the region.

Here is a deeper look at the campaign activity TAG has observed and the actions the team has taken to protect our users over the past few weeks:

APT28 or Fancy Bear, a threat actor attributed to Russia GRU, was observed targeting users in Ukraine with a new variant of malware. The malware, distributed via email attachments inside of password protected zip files (ua_report.zip), is a .Net executable that when executed steals cookies and saved passwords from Chrome, Edge and Firefox browsers. The data is then exfiltrated via email to a compromised email account.

Malware samples:

TAG would like to thank the Yahoo! Paranoids Advanced Cyber Threats Team for their collaboration in this investigation.

Turla, a group TAG attributes to Russia FSB, continues to run campaigns against the Baltics, targeting defense and cybersecurity organizations in the region. Similar to recently observed activity, these campaigns were sent via email and contained a unique link per target that led to a DOCX file hosted on attacker controlled infrastructure. When opened, the DOCX file would attempt to download a unique PNG file from the same attacker controlled domain.

Recently observed Turla domains:

  • wkoinfo.webredirect[.]org
  • jadlactnato.webredirect[.]org

COLDRIVER, a Russian-based threat actor sometimes referred to as Callisto, continues to use Gmail accounts to send credential phishing emails to a variety of Google and non-Google accounts. The targets include government and defense officials, politicians, NGOs and think tanks, and journalists. The group's tactics, techniques and procedures (TTPs) for these campaigns have shifted slightly from including phishing links directly in the email, to also linking to PDFs and/or DOCs hosted on Google Drive and Microsoft One Drive. Within these files is a link to an attacker controlled phishing domain.

These phishing domains have been blocked through Google Safe Browsing – a service that identifies unsafe websites across the web and notifies users and website owners of potential harm.

An example of this technique

An example of this technique

Recently observed COLDRIVER credential phishing domains:

  • cache-dns[.]com
  • docs-shared[.]com
  • documents-forwarding[.]com
  • documents-preview[.]com
  • protection-link[.]online
  • webresources[.]live

Ghostwriter, a Belarusian threat actor, has remained active during the course of the war and recently resumed targeting of Gmail accounts via credential phishing. This campaign, targeting high risk individuals in Ukraine, contained links leading to compromised websites where the first stage phishing page was hosted. If the user clicked continue, they would be redirected to an attacker controlled site that collected the users credentials. There were no accounts compromised from this campaign and Google will alert all targeted users of these attempts through our monthly government-backed attacker warnings.

Both pages from this campaign are shown below.

an example webpage
An example page

In mid-April, TAG detected a Ghostwriter credential phishing campaign targeting Facebook users. The targets, primarily located in Lithuania, were sent links to attacker controlled domains from a domain spoofing the Facebook security team.

Facebook campaign

Recently observed Ghostwriter credential phishing domains and emails:

  • noreply.accountsverify[.]top
  • microsoftonline.email-verify[.]top
  • lt-microsoftgroup.serure-email[.]online
  • facebook.com-validation[.]top
  • lt-meta.com-verification[.]top
  • lt-facebook.com-verification[.]top
  • secure@facebookgroup[.]lt

Curious Gorge, a group TAG attributes to China's PLA SSF, has remained active against government, military, logistics and manufacturing organizations in Ukraine, Russia and Central Asia. In Russia, long running campaigns against multiple government organizations have continued, including the Ministry of Foreign Affairs. Over the past week, TAG identified additional compromises impacting multiple Russian defense contractors and manufacturers and a Russian logistics company.

Protecting Our Users

Upon discovery, all identified websites and domains were added to Safe Browsing to protect users from further exploitation. We also send all targeted Gmail and Workspace users government-backed attacker alerts notifying them of the activity. We encourage any potential targets to enable Google Account Level Enhanced Safe Browsing and ensure that all devices are updated.

The team continues to work around the clock, focusing on the safety and security of our users and the platforms that help them access and share important information. We’ll continue to take action, identify bad actors and share relevant information with others across industry and governments, with the goal of bringing awareness to these issues, protecting users and preventing future attacks. While we are actively monitoring activity related to Ukraine and Russia, we continue to be just as vigilant in relation to other threat actors globally, to ensure that they do not take advantage of everyone’s focus on this region.