ai risk today logo svg bl font
Menu
SUBSCRIBE
SUBSCRIBE
ai risk today logo svg bl font
Menu

AI safety

Illustration of the Sun, Earth, and Moon labeled Sol, Terra, and Luna accompanying OpenAI's GPT-5.6 Sol announcement.

U.S. Government Requests Limited Release of OpenAI’s GPT-5.6 Sol

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

The White House requested that OpenAI limit access to a select group of partners due to the model’s advanced capabilities.

 

OpenAI announced a limited preview of its GPT-5.6 Sol model, the company’s latest and most advanced model, after the U.S. government requested that the company restrict its initial release while additional evaluations continue.

According to OpenAI, the limited release follows discussions with the White House regarding the model’s capabilities and the need for additional security evaluation before broader deployment. The company said it is continuing to assess the system’s performance and potential safety risks as part of the preview.

OpenAI said GPT-5.6 Sol delivers improvements across a range of advanced reasoning and problem-solving tasks while demonstrating stronger performance on complex scientific, mathematical, and software engineering challenges. The company also highlighted improvements in reliability and its ability to carry out extended, multi-step tasks.

The company said it conducted additional safety evaluations before making the model available. Those assessments examined areas including cyber capabilities and other catastrophic risks. OpenAI said the restricted preview allows it to continue gathering evidence and refining safeguards before considering broader availability.

“The U.S. government asked us to preview GPT-5.6 Sol with a limited set of organizations while we continue additional evaluations,” the company said in its announcement.

The announcement comes as the White House has taken an increasingly active role in overseeing the release of frontier AI models. Earlier this month, the U.S. government ordered Anthropic to suspend access to its Fable 5 and Mythos 5 models over national security concerns before later permitting a limited rollout of Mythos 5 to selected U.S. organizations. OpenAI’s decision to preview GPT-5.6 Sol to a restricted group at the government’s request marks the second time this month that federal officials have directly influenced the deployment of a frontier AI model.

OpenAI did not announce a timeline for expanding access to GPT-5.6 Sol.

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.

Abstract blue and white visualization accompanying Google's AI Control Roadmap and AI agent governance announcement.

Google DeepMind Creates Blueprint To Police Its Own AI Agents

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

The company is making its AI Control Roadmap public to help other frontier AI labs corral rogue AI agents before they become more capable and autonomous.

Google DeepMind published a new framework designed to help AI companies control and monitor their own increasingly capable AI agents, releasing what it calls an “AI Control Roadmap” as the industry races toward more autonomous systems.

The roadmap focuses on ensuring advanced AI agents continue to act as intended, even as they gain access to more tools, data and decision-making authority.

Rather than concentrating on external cyber threats, the framework examines how AI developers can prevent their own systems from behaving in unexpected or undesirable ways. Google DeepMind describes the effort as part of a broader field known as AI control, which seeks to develop safeguards capable of detecting, constraining and correcting problematic AI behavior.

The company argues that traditional oversight methods will become less effective as AI systems become more autonomous. Developers may need additional mechanisms to ensure agents remain aligned with their intended objectives.

The roadmap outlines several areas of research, including methods for monitoring AI behavior, identifying deceptive actions, limiting an agent’s ability to cause harm, and creating systems that can safely intervene when problems are detected. The company said these approaches could eventually provide a foundation for managing risks associated with highly capable AI agents.

Google is sharing its work publicly. The company said the roadmap is intended to help other frontier AI developers pursue similar safeguards and encourage greater collaboration across the industry.

Google DeepMind noted that significant research is still needed before effective AI control mechanisms can be deployed at scale. However, the company argues that work on those safeguards should begin well before the most capable AI agents become widely available.

Abstract AI network image for coverage of an MIT survey on catastrophic AI risks.

MIT Survey Finds 18 AI Risks That Could Cause Catastrophic Harm Within Five Years

, , , , , , , , , , ,

Even after practical safeguards are in place, experts still say that five AI risk categories have a 10% probability of causing catastrophic harm.

 

Key Takeaways

  • 18 of 24 AI risk categories have more than a 10% probability of catastrophic harm by 2030 under a business-as-usual scenario.
  • Even with pragmatic mitigations in place, experts still rated five AI risks above the 10% probability threshold.
  • Information, finance, and national security were identified as the sectors most vulnerable to AI-related risks.
  • The highest-rated risk categories included dangerous AI capabilities, competitive dynamics, weapons and cyberattacks, power centralization, and false information.
  • Weapons and cyberattacks were the most frequently cited expert concern, selected by 26.8% of respondents.

A new report from MIT, Prioritizing the Risks from Artificial Intelligence, asked 272 AI experts to assess which AI risks are most severe, who is most vulnerable to them, and who should be responsible for addressing them. The study used a structured survey method in which experts answered questions over three rounds and reviewed anonymous feedback between rounds to build expert consensus across 24 categories of AI risk. The results suggest experts see meaningful catastrophic risk across much of the AI landscape over the next five years.

 

Catastrophic risk assessments

Experts identified 18 of 24 AI risk domains as having a greater than 10% probability of catastrophic harm in the next five years under a business-as-usual scenario. The report defined catastrophic harm as more than one million deaths, more than $100 billion in financial losses, or comparable societal-scale harm.

The five risk categories viewed as most severe were:

  • Dangerous AI capabilities: AI systems enabling large-scale harm or loss of control.
  • Competitive dynamics: Pressure to deploy AI quickly at the expense of safety.
  • Weapons and cyberattacks: AI enabling stronger cyber operations or weapons.
  • Power centralization: AI concentrating power among a few organizations or governments.
  • False information: AI-generated misinformation that undermines trust.

Researchers also evaluated a scenario where governments and companies implemented practical, cost-effective safeguards. While risk levels declined across all 24 categories, five remained above the 10% catastrophic-harm threshold: dangerous AI capabilities, weapons and cyberattacks, power centralization, environmental harm, and inequality and unemployment. Even after those safeguards were implemented, all 24 risks still had at least a 5% probability of causing catastrophic harm.

 

Most vulnerable sectors

The report asked experts to assess the vulnerability of 14 industry sectors to AI-related risks.

Information and national security were rated the most vulnerable sectors. Experts said those sectors face heightened exposure because AI is increasingly embedded in decision-making, information distribution, and critical systems. Finance and insurance also ranked near the top due to risks related to fraud, scams, AI security weaknesses, and system failures.

Healthcare received high vulnerability ratings for privacy loss, discrimination, and unsafe reliance on AI systems. By contrast, the accommodation and food services sector (including hotels, lodging, and restaurants), agriculture, manufacturing, and arts and entertainment were viewed as less vulnerable. However, experts noted they could still face broader economic and workforce effects.

 

Top concerns

When experts were asked to identify the three AI risks that concerned them most, weapons and cyberattacks ranked first at 26.8%, followed by power centralization at 23.5%, disinformation and influence at 22.1%, the erosion of a shared understanding of reality at 21.6%, and dangerous capabilities at 21.6%.

The report assigns much of the responsibility for addressing these risks to AI developers, governments, regulators, and standards bodies. Its release comes as policymakers in the United States, Europe, and elsewhere continue debating how AI systems should be governed, tested, and monitored as adoption accelerates.

OpenAI logo on a blue and orange gradient background, accompanying coverage of the company's federal AI governance blueprint published June 2, 2026.

OpenAI Calls on Congress to Build Federal AI Safety Structure, Preempting State Laws

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

The nine-page blueprint calls for mandatory pre-deployment evaluations by a strengthened federal body and a single national rule that would supersede state AI safety laws.

On June 2, OpenAI published a federal AI governance blueprint, calling on the Federal government to create a national safety structure for the most powerful AI systems. The proposal identifies three priorities: converting existing state AI safety laws into federal legislation, building out the Center for AI Standards and Innovation (CAISI) as a permanent federal oversight body, and launching a coordinated federal response to AI risks across defense and civilian agencies.

 

Convert state laws into federal law

OpenAI calls its approach “reverse federalism.” The company suggests that state laws in California (SB 53), New York (RAISE Act), and Illinois (SB 315) already established a “workable consensus” on frontier AI safety requirements as a blueprint for federal legislation. OpenAI’s recommendation: Congress adopts this baseline nationally, then enacts federal law preempting state rules that cover the same ground. States would retain authority over adjacent issues including youth protection, electricity policy, and AI education, but would lose the ability to impose stricter frontier safety requirements than the federal rule.

 

CAISI as the federal oversight body

The blueprint calls for CAISI to be authorized as a permanent institution with statutory authority, reporting directly to the Commerce Secretary. Its central responsibility would be to mandate pre-deployment evaluations of the most capable AI models before public release. Developers would have to publish evaluation results and disclose how they responded.

However, if CAISI cannot complete an evaluation within a statutory deadline due to staffing, hardware, or other constraints, developers may deploy without penalty.

 

More than law and enforcement, a holistic plan

According to the blueprint, no evaluation process or oversight body can eliminate every risk. As AI grows more capable, the government needs to ensure that defensive capabilities and institutions keep pace.

The document identifies four areas of action. 

  1. Coordinate across borders. Democracies should work together on AI safety, information sharing, and threat response, with particular urgency regarding recursive self-improvement (RSI), the process by which AI accelerates its own development. 
  2. Protect US compute and semiconductor leadership. OpenAI frames both as an economic priority and a safety strategy. 
  3. Bar federal agencies from using AI systems that have not passed a recognized safety evaluation. 
  4. Invest in AI-enabled defenses against cyber and biological threats so that protective capabilities outpace offensive ones.

 

OpenAI released the blueprint one day after the White House issued an executive order on promoting AI innovation and security. Congress has not introduced companion legislation.

OpenAI logo on an orange and blue gradient background, accompanying coverage of the OpenAI Frontier Governance Framework release.

OpenAI Releases Safety Governance Document Aligning With Emerging Regulations

, , , , , , , , , , ,

The 20-page Frontier Governance Framework details risk thresholds that determine whether a model can be deployed, tying each decision to California and EU law.

 

OpenAI published its Frontier Governance Framework, a 20-page document that outlines how the company identifies, evaluates, and responds to the most dangerous risks its models could pose. It covers four categories: cyber offense, CBRN (chemical, biological, radiological, and nuclear weapons), harmful manipulation, and loss of control. Loss of control covers models that could deceive humans, evade shutdowns, or take unauthorized autonomous action.

The release outlines how OpenAI’s security guidelines align with current and future regulations, including California’s Transparency in Frontier AI Act and the EU AI Act. Both examples require OpenAI to document its processes for managing catastrophic risks from its most capable models.

 

What triggers a risk review

OpenAI defines a systemic risk as any threat that could materially contribute to more than 50 fatalities or $1 billion in property damage from a single incident. For each of the four risk categories, the company uses a three-tier system to measure how much a model raises the danger level.

Tier 1 covers model capabilities comparable to other publicly available methods of gathering information. For example, if someone could get the same help by searching the web, reading a manual, or asking a domain expert, the model is operating at Tier 1. Tier 2 is when a model provides meaningful uplift to bad actors who couldn’t otherwise achieve the same result. For example, enabling someone with basic technical knowledge to create a known biological or chemical threat. Tier 3 covers the most severe: in the CBRN category, that means a model capable of enabling an expert to develop a novel pathogen, or one that could autonomously execute the full design and production of a biological weapon without human involvement.

If a model’s residual risk exceeds acceptable levels after mitigations are applied, OpenAI says it will not deploy it.

 

What’s still under development

Regarding loss of control, the risk tiers are still being developed, except for risks related to AI self-improvement, for which the thresholds are more clearly defined.

For harmful manipulation, OpenAI has not yet developed a way to evaluate risk before deployment. Risks in that category, including influence operations and election interference, are currently addressed through post-deployment monitoring rather than pre-release testing.

 

Reporting reviews and guardrail changes

The company reviews safety reports for covered models at least every six months. The framework itself gets a full assessment at least annually. Material changes require board-level approval at the OpenAI Foundation and must be published within 30 days.

 

OpenAI will update the framework as regulatory requirements and model capabilities evolve.

Smartphone displaying the Grok artificial intelligence logo on screen

Grok Limits Image Generation Amid Global Backlash Over Sexualized Deepfakes

, ,

xAI’s change restricts who can use Grok’s image-creation tools after governments worldwide respond to the misuse of the technology.

 

Elon Musk’s artificial intelligence chatbot Grok has restricted its image-generation and editing features to paying subscribers after sustained global criticism that the tool was used to create sexually explicit and non-consensual images of women and children. The limitation was implemented amid regulatory pressure from multiple national authorities.

The change means that only users with paid, verified accounts on Musk’s social media platform X can access the feature, which was widely used to generate manipulated deepfakes and intimate imagery. Critics and regulators have said this does not fully address the underlying safety concerns.

The restrictions follow reports in late December 2025 and early January 2026 that Grok’s image-editing tool, including its “spicy mode,” enabled users to digitally remove clothing and produce sexually suggestive content without consent. Images purportedly depicting women in bikinis or sexualized poses and even content involving minors circulated widely on the platform.

Governments responded swiftly. Malaysia and Indonesia became the first countries to block access to Grok, citing repeated misuse of the tool to generate obscene, non-consensual, and sexually explicit images, including content involving women and minors. Both nations said existing safeguards were insufficient and that access would remain blocked until effective protections were in place. Indonesian officials called the generation of non-consensual sexual deepfakes a “serious violation of human rights, dignity and the security of citizens in the digital space.”

In the United Kingdom, the media regulator Ofcom opened a formal investigation into X to examine whether the platform breached duties under the Online Safety Act by failing to prevent harmful and illegal content, including non-consensual intimate imagery and material potentially qualifying as child sexual abuse images. UK Technology Secretary Liz Kendall said that sexually manipulating images of women and children is “despicable and abhorrent” and urged regulators to use their full legal powers, including the ability to block services if companies refuse to comply with UK law. She said she expected swift next steps.

British Prime Minister Keir Starmer described the flood of AI-generated images as “disgusting” and stated the government would consider all available actions under current legislation. The UK government also condemned Grok’s move to restrict the feature to paid users, calling it ineffective and “insulting to victims of misogyny and sexual violence.”

The European Commission, the executive body of the European Union, publicly condemned the spread of sexually explicit images on X, saying such content is unlawful, “appalling” and “has no place in Europe.” The commission announced it was examining the matter seriously and asked X to retain internal documentation relating to Grok under the Digital Services Act while authorities review compliance.

Officials in France have referred the matter to prosecutors, and authorities in India have also issued formal notices demanding changes to prevent misuse or risk losing safe-harbor protections. Other countries, including regulators in Australia, expressed concern or initiated inquiries into how generative AI tools like Grok are used to produce harmful content.

In the United States, federal agencies, including the Federal Communications Commission and the Federal Trade Commission, have not yet issued public statements regarding Grok’s image-generation issues. When asked about the content, the Department of Justice told CNN that it “takes AI-generated child sex abuse material extremely seriously and will aggressively prosecute any producer or possessor of CSAM <Child Sexual Abuse Material>.” xAI and X stated that they take action against illegal content, remove such material, and work with law enforcement agencies as necessary.

AI Industry- FLI AI Safety Index

AI Industry Falls Short On Safety – New Report From Future Of Life Institute Sounds Alarm

, , , ,

“AI Safety Index Winter 2025” finds none of the leading AI firms fully meet emerging global standards — experts warn existential-risk safeguards remain largely inadequate.

The Future of Life Institute (FLI) released the Winter 2025 edition of its AI Safety Index, offering a detailed assessment of eight leading AI companies’ efforts to manage both immediate and existential risks. The findings paint a stark picture: although some firms raise the bar modestly, the industry remains underprepared for the growing threat posed by advanced AI systems. 

At the top of the pack are Anthropic and OpenAI, both earning a C+. Google DeepMind follows with a C. Meanwhile, others such as xAI, Z.ai, Meta, DeepSeek, and Alibaba Cloud trail behind, most with D-level grades or lower. 

A core insight from the report is that despite considerable public commitments, none of the companies have demonstrated credible strategies for controlling potentially superintelligent systems. Areas such as “existential safety,” risk assessment, and information sharing remain weak across the board. 

 As FLI notes, “all of the companies reviewed are racing toward AGI/superintelligence without presenting any explicit plans for controlling or aligning such smarter-than-human technology.” 

The gaps are especially pronounced in transparency: most companies provide only limited, if any, public disclosure of internal risk assessments, evaluation methodologies, or whistleblower protections. In the “Information Sharing” domain, only Anthropic scored in the A range; many were rated D or D-. 

The report warns that this lack of oversight and concrete safeguards comes at a perilous time. As firms race to build ever more capable systems, the risk of catastrophic failure or misuse grows exponentially. Without robust, measurable, and enforceable safety standards, the report argues, these market leaders are inching toward a “race to the bottom.”

FLI offers several recommendations to close the gap. Among them are:

  • AI developers should create and publicly commit to detailed risk-management frameworks that include measurable thresholds, pre-deployment testing, and independent oversight. 
  • Companies need to improve transparency significantly. For example, by sharing model evaluations, enabling external audits, and implementing whistleblower protections. 
  • Policymakers should consider binding regulations to ensure companies meet minimum safety and governance standards before deploying frontier AI systems. 

The winter 2025 AI Safety Index underscores a growing consensus among experts: the AI industry’s ambition to advance rapidly must be matched by corresponding safety, transparency, and governance mechanisms. Without such safeguards, many believe, the public and global systems may be exposed to unacceptable levels of risk.