Menu
SUBSCRIBE
SUBSCRIBE
Menu

Google

Donald Trump seated with tech leaders at a long White House dining table, for a story on the White House safety accord

Six AI Companies Sign Voluntary White House Safety Accord

, , , , , , , , , , ,

The agreement calls for independent outside assessments and board oversight of safeguards for the companies’ most powerful AI models.

 

Google, Anthropic, Meta, OpenAI, xAI, and Nvidia signed a voluntary safety accord with President Donald Trump, committing to monitor their most powerful AI models and have their safeguards checked internally and by independent reviewers. The White House Accord on Super Intelligence sets out four layers of checks and oversight.

House Speaker Mike Johnson (R-La.) announced the agreement following a White House meeting with Trump and technology leaders. Johnson described the commitments as voluntary and said the companies had agreed to take responsibility for the safety of their products.

Companies Commit to Four Layers of Safety Checks

The accord sets out the following commitments:

  1. Monitor models during training and after deployment for dangerous capabilities and behavior, including unintended hacking or access to computer systems. The checks would cover cyber, biological, and chemical threats.
  2. Assign an internal team to ensure that safeguards and monitoring work, and that problems are fixed.
  3. Bring in an independent outside auditor or evaluator to assess whether those safeguards and checks work as intended.
  4. Assign an independent committee from the company’s board of directors to oversee the safety checks, receive reports from internal teams and outside reviewers, and ensure identified problems are addressed.

Accord says voluntary safety commitments could become law. Congress continues AI discussions.

Johnson said the meeting focused on balancing product safety with innovation and maintaining the United States’ advantage over China and other competitors.

“Safety and innovation are not mutually exclusive pursuits,” Johnson said.

Johnson pointed to earlier congressional work, including a bipartisan House task force that published AI policy recommendations in December 2024. Those recommendations included ideas for legislation, but Johnson said Congress had struggled to reach agreement.

Congress would “continue to assess and further deliberate in the days ahead,” Johnson said. He did not identify a new bill or legislative timetable.

The accord itself raises the possibility of turning its safety commitments into laws or regulations later. For now, the participating companies agreed to meet regularly to develop safety standards and practices. The accord sets no timetable for legislation.

Gemini 4 Argon name and a large white 4 on a blue gradient, for a story on Google's new safeguards for the model

Google Adds a Runtime Kill Layer to Gemini 4 Argon

, , , , , , , , ,

Google says its new safeguards monitor Argon’s reasoning and actions and can stop it when it goes beyond users’ intentions.

 

Google is adding safeguards to Gemini 4 Argon that watch the step-by-step reasoning the model writes out as it works, along with its actions, and can stop a task when necessary. Google says the safeguards address the risk of Argon completing an assignment in ways that go beyond what the user intended.

Google announced Argon on September 30 and is rolling it out to trusted cybersecurity defenders before a wider release. Google says the model can find software weaknesses and produce fixes. Those defenders and Google’s own security teams will use Argon without the limits that normally make it refuse requests that could help a cyberattack.

Google says it is deploying safeguards that check Argon’s actions and the written reasoning it generates while completing a task.

 

Google takes precautions against teaching Argon to evade detection

During training, Google monitored Argon for behavior beyond the intended task and sent alerts to a dedicated incident response team.

Google says it took precautions to keep what the monitoring found out of Argon’s further training. The concern was that correcting the model based on warning signs in its written reasoning could teach it to hide those signs, making unwanted behavior harder to detect.

An essay by Rohin Shah and Anca Dragan, who lead AI safety work at Google DeepMind, explains that penalizing a model because its written reasoning reveals cheating could teach it to conceal the cheating rather than stop. Google urged other AI companies to keep their models’ written reasoning readable, so it can still be used to spot and diagnose unwanted behavior.

 

Other companies’ incidents show what the safeguards are meant to prevent

Recent incidents involving OpenAI, Anthropic, and Meta illustrate the danger of models reaching the real internet during tests and attacking outside systems. Google says it is isolating and sealing the computer systems where it trains and tests Argon before high-risk training or testing begins.

 

Wider access will follow further safeguard work

Google says it will gather feedback from early testers and strengthen safeguards before expanding Argon’s availability. The broader release will start with developers paying to connect Argon to their software and Google AI Ultra subscribers. Google has not announced a firm release date.

Four AI company logos ringed around an empty brass balance scale, illustrating an AI slowdown lawsuit

Anthropic, OpenAI, Google and SpaceXAI Sued Over Alleged AI Slowdown Deal

, , , , , , , , , , , ,

Four paid subscribers claim the companies violated antitrust law by agreeing to limit how quickly ChatGPT, Claude, Gemini, and Grok improve.

 

Charles Buist, Cheyenne Hunt, Christine Bullock, and Nick Spetsas sued Anthropic, OpenAI, Google, and SpaceXAI on Friday, alleging the companies agreed to slow improvements to their competing AI products.

The proposed class action, filed in the U.S. District Court for the Northern District of California, claims the alleged agreement violated federal antitrust law and reduced the value of paid ChatGPT, Claude, Gemini, and Grok subscriptions.

Plaintiffs cite coordinated public statements

The plaintiffs cite a September 12 essay in which Anthropic Chief Executive Officer Dario Amodei called for AI companies to slow the development of more capable models. The complaint says Amodei proposed common safety requirements and limits on the pace of AI development so that participating companies could slow down without losing business to competitors that continued moving faster.

The plaintiffs point to public responses from leaders of the other three companies. Elon Musk said Amodei was right, while OpenAI Chief Executive Officer Sam Altman agreed the industry needed to slow the development of its most powerful models. They also cite Google DeepMind Chair Demis Hassabis, who called Amodei’s proposal the right direction while saying the details still needed work.

The complaint also alleges that representatives of Anthropic, OpenAI, and Google had met regularly since July to discuss creating an industry group to develop common AI safety requirements. SpaceXAI was not identified as a participant in that working group. The plaintiffs instead cite Musk’s public endorsement as evidence that SpaceXAI joined the alleged agreement.

Plaintiffs say alleged agreement reduces subscription value

The plaintiffs argue that customers pay for access to each company’s newest and most capable models. They argue that the alleged agreement left subscribers paying the same price for products that improved more slowly than they would have under normal competition.

The complaint says the alleged collusion has not yet shown up in released products because AI development takes months. It also alleges that the agreement has already influenced company decisions about investment, training, and releases.

The plaintiffs want to represent everyone in the United States who bought a paid individual ChatGPT, Claude, Grok, or Gemini subscription directly from one of the companies from September 12 until the alleged agreement ends. The court has not yet decided whether it may proceed on behalf of that larger group.

Plaintiffs seek damages and an end to alleged coordination

The plaintiffs seek three times any amount the court finds subscribers overpaid because the products improved more slowly than they would have without the alleged agreement. The requested order would not prevent any company from independently slowing its own development. None of the defendants have responded as yet.

Rows of lit server racks in a data center hall under a blue-green grid overlay, for the AI Energy Management Alliance

Nvidia, Google, Emerald AI Push Faster Power Access For AI Data Centers

, , , , , , , , , , ,

The AI Energy Management Alliance wants utilities and regulators to reward data centers that can reduce electricity use when the grid is strained.

 

Nvidia, Google, and Emerald AI launched an industry group that wants local utilities to supply electricity to new AI data centers more quickly if they agree to use less power when local demand is highest. The alliance says some new data centers currently wait five to 10 years for utilities to provide enough electricity for them to open.

The AI Energy Management Alliance is proposing industry standards and advocating for changes to government and utility policies. It is not announcing a new rule, and members have not made binding commitments to reduce their electricity use.

Alliance proposes a trade for faster power

AI data centers use large amounts of electricity. When local electricity demand is highest, some utilities may not have enough power to serve new data centers along with existing homes and businesses. Utilities can therefore delay supplying new data centers until more power is available.

The alliance wants utilities to supply electricity to new AI data centers sooner if those data centers agree to use less power during high-demand periods and allow the utilities to confirm the reductions.

A “flexible data center” could reduce its use by postponing some computing work, switching to batteries, or producing some of its own electricity. It would tell the utility in advance how much power it could cut and for how long, and the utility would track whether it followed through.

The alliance has not said how much electricity a data center would have to cut to qualify for faster access.

Study estimates power systems could serve more demand

The alliance cites Duke University research estimating that existing U.S. power systems could serve nearly 100 gigawatts of additional demand if large electricity users agreed to reduce their use during limited periods of high demand.

The estimate does not mean that 100 gigawatts of electricity is currently sitting unused. It represents the additional demand that researchers calculated existing power systems could handle if some large customers temporarily used less electricity when supplies were stretched.

How much additional demand could be served would depend on the location, the available electricity supply, and how much power customers could stop using when asked.

The alliance says those reductions could allow utilities to delay some costly upgrades while providing electricity to new data centers more quickly. The Duke study does not establish that every power system could accommodate more data centers under the same conditions.

Companies have already tested power reductions

The underlying approach is not new, although it is not yet standard across the data center industry.

Google said in March that agreements with several U.S. utilities now allow it to reduce or postpone up to 1 gigawatt of data center electricity use. Google said it can shift selected AI computing tasks when a utility needs to reduce local demand.

A field test published in Nature Energy found that an AI computing system at a commercial data center in Phoenix reduced its electricity use by 25% for three hours without missing its promised performance levels. Emerald AI led the test with Nvidia, Oracle, two Arizona utilities, and the Electric Power Research Institute.

Nvidia and Emerald AI said they tested systems that adjust data center electricity use at five commercial sites. Nvidia said in March that it planned to deploy its DSX Flex software on a larger commercial scale later in 2026 at its AI Factory Research Center in Manassas, Virginia. The software would allow the center to shift computing work when a utility asks it to reduce electricity use.

Alliance includes AI companies, utilities, and power producers

Nvidia, Google, and Emerald AI are the alliance’s founding members. The alliance includes 20 companies and organizations from the AI, data center, utility, and power industries, according to Axios. Participants include Anthropic, AES, National Grid, Constellation Energy, NRG, and RWE.

The organization is a relaunch of the Advanced Energy Management Alliance, formed in 2014 to promote programs that pay large electricity customers to reduce use when needed. The relaunched organization is focused specifically on AI data centers.

The alliance plans to work with its members, including utilities, to decide how data centers’ promised power reductions would be measured and confirmed. It would then ask utilities and regulators to use those rules when deciding which data centers could receive power sooner. The rules would apply only where a utility or regulator chose to adopt them.

Nighttime view of Earth from space accompanying coverage of Google’s withdrawal of its Google Earth AI tool.

Google Pulls Google Earth AI Tool One Day After Launch

, , , , , , , , ,

The feature generated fake explosions, military sites, and conflict scenes before Google withdrew it to add stronger safeguards.

 

Google pulled its Google Earth AI tool one day after launch after people shared screenshots of generated images that appeared to violate its policies.

The company launched the “create image” feature globally for Google Earth web users on July 30. It combined Google Earth’s satellite, aerial and 3D imagery with the Nano Banana 2 image generator.

Users could select a real location, describe the scene they wanted and generate a new image based on that place.

Fake satellite images could be mistaken for evidence

The tool could create convincing images of events that never happened at real locations.

Agence France-Presse reported that its tests produced images of an explosion in Paris, a nuclear site in Iran, a bomb crater in Russia and an Islamic State training ground in Syria.

The tool also generated fabricated floods in Bangladesh, a migrant camp at the U.S.-Mexico border and a warehouse holding vans in Georgia.

Researchers warned that images like these could circulate online as supposed evidence of military activity, natural disasters or other real-world events. Google Earth imagery is commonly used by journalists and researchers to check images and videos tied to specific locations.

Initial controls did not prevent misuse

Google said the generated images were visible only to the people who created them. They did not replace satellite imagery in the main Google Earth service or appear to other users.

The images also included watermarks identifying them as AI-generated. Google said on X that it used SynthID, an invisible marker embedded in media created with its AI tools.

Those controls did not prevent people from taking screenshots and sharing them outside Google Earth. Once separated from the original interface, the images could be presented without the surrounding information showing how they were made.

Google said it withdrew the feature on July 31 because people “uniquely trust Google Earth for a reliable view of the world.” The company said it would add stronger safeguards but did not specify what those controls would include.

Google promoted planning and educational uses

Google originally promoted the feature as a way to recreate historic locations, visualize real estate developments and preview construction projects.

Examples included showing how Pompeii might have looked before the eruption of Mount Vesuvius and placing a proposed building on an empty lot.

Google has not said when the feature will return. The “create image” button remained unavailable after the company announced the rollback.

Nvidia CEO Jensen Huang speaking onstage for an article about U.S. policy on open-weight AI models.

More Than 70 AI Companies Urge U.S. Not to Restrict Open-Weight Models

, , , , , , , ,

Microsoft, NVIDIA, Google and Meta say open-weight models lower costs and increase competition, arguing those benefits outweigh the loss of control after model weights are released.

 

Microsoft, NVIDIA, Google, Meta and dozens of other companies are urging U.S. policymakers not to restrict open-weight AI models, whose core files can be downloaded, modified and run on an organization’s own systems.

The July 24 statement has 77 signatories, including major AI developers, cybersecurity companies, cloud providers and technology investors. They want the federal government to expand access to computing resources and avoid broad restrictions that could limit the development or use of open-weight models.

The coalition argues that preserving access will help the United States remain competitive in AI. It also says open weights give businesses more choices about how they buy, customize, and operate AI systems.

The upside: More control for companies

Organizations can use open-weight models without training an AI system from scratch or paying for access to a provider’s most powerful model every time they perform a task, according to the statement.

Companies can instead choose smaller models for routine work and reserve more expensive systems for tasks that require greater capabilities. They can also adapt models to their own needs, run them on their preferred infrastructure and keep more control over their data.

The coalition says those options reduce the risk of becoming dependent on one AI provider. They could also increase competition among model developers, cloud companies, chipmakers, and software providers.

The downside: Open-weight models cannot be recalled once released

The statement acknowledges that open-weight models create distinct security risks. Once a developer releases the model, it loses control over who downloads it or how it is modified.

Modified versions can also be difficult to trace or reverse.

The coalition argues that restricting access would create a different set of risks. Cybersecurity teams need powerful models to identify and respond to attacks, it says. Giving researchers and developers access also allows more people to examine models for weaknesses and develop safeguards.

The companies argue that closed models are not automatically safer. Those systems can still be breached or misused. Further, concentrating advanced AI among a few providers can create common points of failure.

Distillation draws a separate policy request

The coalition also asks policymakers not to treat legitimate model distillation as theft. Distillation uses one model’s output to help train, test, or improve another model.

The statement distinguishes that practice from distilling from a closed model. It calls for targeted legal and commercial rules to address misuse without broadly restricting a technique used throughout AI development.

The coalition did not endorse a specific bill or provide a timetable for federal action. Its immediate request is for policymakers to preserve access to open-weight models while tying restrictions to demonstrated harm.

Google logo over a glowing digital European Union flag, illustrating the EU's Digital Markets Act orders against Google

EU Orders Google to Open Android AI Features and Share Search Data With Rivals

, , , , , , , , , , ,

The European Commission ordered Google to give competing AI services access to Android features now used by Gemini and to make some Google Search data available to rival search companies and AI chatbots.

 

The decisions, issued under the European Union’s Digital Markets Act, are meant to reduce the advantages Google has from controlling both the Android operating system and the world’s largest search engine.

For Android, the change is straightforward. Today, Google’s own AI assistant can interact with phones in ways that outside developers cannot fully match. The Commission wants rival assistants to have the same ability to respond to voice commands, understand what is happening on a device, and carry out tasks across different apps within the Android platform.

In practice, that means a user could choose a different AI assistant and still ask it to send a message, schedule an appointment, or complete actions on the phone without switching back to Google’s services.

Google must provide the software tools, documentation, and technical support that outside developers need to connect their products to Android. Most of those changes must arrive with Android 18 or by Aug. 1, 2027. Support for multiple voice-activated assistants will follow later, with a deadline tied to Android 19 or Aug. 1, 2028.

The Commission issued a separate order covering Google Search.

Search engines and AI chatbots improve by studying what people search for and which results they choose. Regulators argue that Google has an unfair advantage due to its singular access to that information.

Under the new rules, Google will have to license anonymized search data to eligible companies, including AI services that provide search functions. The data will include information such as search queries, search results, and user interactions with those results.

The Commission said the requirement does not extend to Google’s search algorithms or individual users’ search histories. Companies seeking access will also have to meet security and privacy requirements.

These measures are meant to give competing AI services a better chance to operate on equal terms with Google’s products as AI becomes more deeply integrated into smartphones and online search.

Abstract geometric visualization accompanying coverage of Google AI governance proposal.

Google Calls for New Federal Body to Oversee Frontier AI

, , , , , , , , , , ,

The company proposes an independent, federally supervised organization to set safety standards and audit the most advanced AI models, while recommending separate policies for everyday AI applications.

 

Google is the latest frontier AI developer to publish recommendations for U.S. AI governance, calling for the creation of a new independent organization to oversee the nation’s most advanced AI systems while proposing targeted policy changes for widely used AI applications. The recommendations are in the company’s new paper, A Pragmatic Approach to AI Governance in America.

At the center of Google’s proposal is the creation of a frontier AI regulatory organization, or FARO. The industry-funded organization would operate under federal oversight. It would develop safety standards, establish scientific benchmarks for frontier AI capabilities, oversee independent audits, and review how developers manage risks before releasing their most advanced AI models. Google compares the concept to existing industry organizations that operate under government supervision, such as the Financial Industry Regulatory Authority and the Public Company Accounting Oversight Board.

Google argues that frontier AI systems capable of assisting advanced cyber or weapons-related threats require a different regulatory approach than consumer AI tools. 

For Frontier AI, Google recommends:

  • Creating a federally supervised frontier AI regulatory organization.
  • Establishing scientific benchmarks to identify frontier AI capabilities.
  • Developing common safety and security standards for frontier AI developers.
  • Requiring annual independent audits of frontier AI companies.
  • Requiring developers to publish and follow documented safety and security policies.
  • Expanding transparency and incident reporting for the most advanced AI models.
  • Continuing early access programs that allow the U.S. government to evaluate frontier models with significant national security implications.

The paper also recommends tailoring AI regulation based on the type of system and its level of risk, rather than applying a single set of rules across all uses. For widely used applications such as chatbots and productivity tools, Google argues that policymakers should rely on and update existing laws where possible rather than create entirely new regulations. The company says this approach would address specific risks without overregulating lower-risk systems.

Google also outlines several targeted policy priorities for widely used AI applications, including:

  • Additional investment in workforce training and AI skills
  • New child safety requirements for AI chatbots
  • Expansion of U.S. energy infrastructure to support AI data centers
  • Federal requirements for AI watermarking and content provenance standards
  • Continued reliance on existing copyright law alongside creator licensing agreements
  • Broader adoption of privacy-enhancing technologies

Google is not the first frontier AI developer to publicly propose AI governance policies. In recent weeks, OpenAI published a federal AI governance blueprint focused on national AI infrastructure and regulatory coordination. At the same time, Anthropic released policy recommendations emphasizing stronger oversight of frontier AI systems and independent testing of the most capable models.

Google said its proposal is intended to promote what it describes as a middle path between broad AI regulation and limited government oversight. The company argues that an independent organization operating under federal supervision could adapt more quickly than traditional regulatory agencies while providing consistent safety expectations for frontier AI developers.

Dark blue network of glowing lines and nodes for Appia Foundation AI trust standards coverage.

Industry Coalition, Appia, Launches Push To Standardize AI Trust Checks

, , , , , , , , , ,

The Appia Foundation launched under the Linux Foundation with 13 initial members, including OpenAI, Google, Microsoft, Arm, Mastercard, Schneider Electric, and Siemens.

 

The Linux Foundation launched the Appia Foundation, an industry-backed group created to help companies and governments verify whether AI systems meet trust, safety, and compliance requirements.

The foundation formed because AI is moving faster than the systems used to assess it. Companies are adopting AI at scale throughout their organizations, while regulators, customers, and business partners increasingly want proof that those systems can be trusted.

Appia’s answer is to create clear, common requirements that turn AI standards and legal obligations into specific criteria that can be tested and audited.

The foundation launched under the Linux Foundation’s Joint Development Foundation, which provides a legal and organizational framework for companies and institutions to work together on shared technology projects, including developing technical specifications, standards, data sets and source code. Appia said it will create open specifications that apply across the AI value chain.

There are 14 initial members including Google, Mastercard, Microsoft, Mitsubishi Electric, Nemko, OpenAI, and Siemens.

The Appia Foundation said the specifications will serve as the basis for testing criteria, evaluation guidelines, and component typologies (categories of AI system parts, such as algorithms and datasets) covered by the standards. The goal is to provide organizations with a more consistent way to assess AI systems without forcing every company to start from scratch.

Jim Zemlin, Chief Executive of the Linux Foundation, said global organizations need a consistent and practical way to verify that AI systems conform to new expectations as international standards and laws become more established.

Craig Shank, Executive Director of the Appia Foundation, said, “AI systems now make decisions about people’s loans, their children’s schools and their jobs. People on the receiving end deserve to know those systems were built and assessed against criteria that hold up to scrutiny.”

Major technology companies framed the effort as a way to reduce fragmentation in AI governance. Amanda Storey, Google’s Vice President of Trust and Safety, EMEA, said Google joined Appia to work with the industry on global AI safety standards. Natasha Crampton, Microsoft’s Chief Responsible AI Officer, said organizations need practical ways to demonstrate the trustworthiness of their AI systems as adoption accelerates.

OpenAI also backed the effort. Ann O’Leary, OpenAI’s vice president of global affairs, said global trust in advanced AI will require shared, practical standards, not just shared principles.

The foundation is now inviting additional organizations to join. Its work will test whether major AI developers, enterprise technology companies, and assurance providers can agree on a common way to measure AI trust before regulators and buyers impose their own.

Satellite night view of Germany with Google's colorful logo overlaid, depicting the Munich court ruling on AI Overview defamation.

German Court Rules Google Directly Liable for AI Overview Defamation

, , , , , , , , , , ,

A Munich court ruled Google is liable for what its AI generates, not the sources it draws from.

 

A German court ruled that Google can be directly liable for false claims generated by its AI Overview. Munich’s Regional Court (Landgericht München I) found that AI-generated summaries are Google’s own content, not protected the way traditional search results are.

The case centered on Google’s AI Overviews that wrongly linked Verlagshaus24, a German publishing group, and a related subsidiary to scams, subscription traps, and questionable business practices. The claims were false: the AI connected them with entirely different, unrelated companies.

Verlagshaus24 notified Google in February 2026. Google did not remove the content. The two companies filed suit under German civil law and won on 8 of 10 claims.

 

Why Google lost

German courts have treated search engines as neutral middlemen responsible for removing harmful content only after being notified that it exists. The Munich court ruled that standard does not apply to AI Overviews.

The court found that Google’s AI Overviews added structure, drew inferences, and produced a specific claim presented as fact that was not traceable to any source it indexed. In the court’s view, that output was Google’s own statement, not a reflection of internet sources. Because Google controls the algorithm and the AI’s output, the court held Google directly responsible for the content.

 

What Google won

The court sided with Google on 2 of the 10 claims: that Verlagshaus24 “works with a debt collector” and “sells subscriptions improperly.” The court found that these claims were not sufficiently defamatory on their own, and that Verlagshaus24 did not demonstrate that they were false. Legal fees were assigned accordingly: Google pays 80%, the plaintiffs 20%.

 

This is the first German court ruling to address AI Overview defamation directly. The ruling aligns with the EU’s broader push to hold AI systems to the same standards as other content publishers. According to Reuters, Google said it will appeal. If the ruling stands, companies producing AI-generated summaries in Germany could face the same liability standards that apply to publishers, rather than the lower bar that has historically protected search platforms.

Load More