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

Anthropic

Vintage butterfly S illustration on a laptop screen amid a blue digital network, representing Anthropic Mythos 5 limited release.

U.S. Allows Limited Release Of Anthropic’s Mythos 5

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

Commerce Secretary Howard Lutnick authorized limited access to Anthropic’s cybersecurity-focused Mythos 5 model for approved organizations while broader restrictions remain in place.

The U.S. Commerce Department approved the limited release of Anthropic’s Mythos 5 AI model to select organizations, while continuing to restrict access to its consumer-focused Fable 5 model.

According to a letter from Commerce Secretary Howard Lutnick reviewed by Reuters, the authorization allows Anthropic to give access to Mythos 5 to a defined group of trusted U.S. organizations following the government’s earlier suspension of both models. Axios, which also reviewed the letter, reported the decision followed what the department described as “significant progress” by Anthropic in implementing appropriate safeguards.

According to Reuters and Axios, the letter includes five key provisions:

  • Limited access to Mythos 5 is restored. Anthropic may now provide the model to approved trusted organizations under the revised authorization.
  • Access is limited to approved organizations. Reuters reported that more than 100 organizations, including many Fortune 500 companies and critical infrastructure partners, are eligible under the authorization.
  • Foreign-national employees may use the model. Employees of approved organizations may access Mythos 5 without requiring additional export licensing, according to Reuters.
  • Commerce retains the authority to modify the decision. Reuters and Axios reported that the department reserved the right to revise or withdraw the authorization if circumstances change.

Earlier this month, the Commerce Department directed Anthropic to suspend access to Mythos 5 and Fable 5 after national security concerns emerged following a reported jailbreak involving the models.

The White House decision follows another recent example of government involvement in frontier AI deployments. On the same day, OpenAI announced it would limit the release of its new GPT-5.6 Sol model at the request of the U.S. government.

Anthropic said it is working to restore access to Mythos 5 for approved organizations while continuing discussions with the U.S. government regarding the broader availability of its frontier AI models.

Anthropic CEO Dario Amodei standing before the White House, representing the company's cybersecurity collaboration with the U.S. government through Project Glasswing

U.S. Government Used Anthropic’s Mythos To Find Classified System Vulnerabilities Within Hours

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

The government participates in Project Glasswing, which gives vetted partners access to Mythos, the Anthropic model that has found more than 10,000 software vulnerabilities since April.

 

The U.S. government used Anthropic’s Mythos AI models to identify vulnerabilities in classified systems within hours, according to the Associated Press. It is the most sensitive disclosed application yet of Project Glasswing, Anthropic’s restricted program that gives vetted organizations access to its most powerful AI models to find software vulnerabilities before attackers do.

It is unclear whether the government used Mythos Preview, the original version launched in April, or Mythos 5, the latest version. The U.S. government suspended access to Mythos 5 on June 12 after citing a jailbreak identified in Fable 5, the consumer version of the same model, and halted access to both on security grounds.

Anthropic launched Project Glasswing, concurrently with Mythos, after determining that Mythos was too effective at finding vulnerabilities to release publicly. The company restricted access to a short list of vetted technology and cybersecurity companies, along with select government organizations. Partners found more than 10,000 high- and critical-severity software vulnerabilities in the program’s first weeks.

The government and Anthropic continue to work together despite the ongoing conflict between the parties. In March, the Department of Defense designated Anthropic a supply chain risk after Anthropic requested assurances its models would not be used for mass domestic surveillance or fully autonomous weapons. Anthropic challenged the designation in federal court and secured a temporary injunction. In June, the government separately ordered Anthropic to suspend access to Fable 5 and Mythos 5 due to concerns about access by foreign nationals. Anthropic complied while publicly disputing the evidence behind the order. The lawsuit remains active.

The White House held direct talks with Anthropic CEO Dario Amodei earlier this year about cybersecurity and shared AI priorities. Anthropic said it plans to expand Glasswing’s membership.

Composite image of Trump and leading AI CEOs above Évian-les-Bains during G7 AI governance discussions.

Trump, Frontier AI CEOs Discuss Global AI Governance at G7

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

President Donald Trump met with leaders of the world’s most influential AI companies during a closed-door discussion that included international AI governance and coordination.

 

President Donald Trump met with leading AI executives, including OpenAI CEO Sam Altman, Anthropic CEO Dario Amodei, and Google DeepMind CEO Demis Hassabis, during a closed-door AI governance discussion at the G7 Summit in Évian-les-Bains, France, on June 17.

According to reporting from Axios, Reuters, and other media outlets, international AI governance and coordination were among the topics discussed during the session. Reuters reported that participants addressed AI governance, international cooperation, access to advanced AI models, and trusted AI partnerships.

Following the meeting, OpenAI Head of Global Affairs Chris Lehane said that governments and AI companies discussed the possibility of creating a forum for democratic countries to work together on AI safety standards.“There really is a coalescing” among countries and AI labs around the idea of establishing “a forum or a space for the different democratic countries to be able to work together to ultimately see if there’s a way to establish some type of AI safety standards,” Lehane told reporters, according to Axios.

Lehane said the United States would play a leading role in such an effort and argued that developing common standards could help preserve access to advanced AI systems. His comments echoed a similar OpenAI proposal published last month that called for a U.S.-led international AI oversight group that included China.

He also said participants discussed children’s online safety and ways to help ensure that AI technology remains broadly available worldwide.

Trump described the discussion as an “excellent” meeting during a press conference following the summit.

“It’s going to be the biggest thing ever,” Trump said. “We have to be very careful with it. It’s both great and could be bad. We have to be careful with it, but we’re leading China. We’re leading the world on that.”

No formal proposal, governance framework, treaty, or regulatory initiative emerged from the meeting.

Butterfly and moth specimens forming the numeral five under a cybersecurity grid, for freefable.org's letter to lift the Fable and Mythos ban.

130+ Technology Leaders Demand U.S. Lift Fable and Mythos Suspension

, , , , , , , , , , ,

freefable.org, formed in response to the June 12 directive, published an open letter arguing the ban disarms defenders without restricting what attackers can already use in other models.

 

More than 130 information security executives, researchers, and academics published an open letter on June 14, formally calling on the U.S. government to lift the Fable and Mythos suspension imposed two days earlier.

The letter was published at freefable.org, an organization formed specifically in response to the directive that prompted Anthropic to cut access for all users on June 12. It is addressed to Commerce Secretary Howard Lutnick and National Cyber Director Sean Cairncross. Notable signers include Bruce Schneier, one of the most widely cited voices in computer security, affiliated with Harvard University and the University of Toronto; Philip Zimmermann, creator of the PGP encryption standard; and Vinh Nguyen, former Chief Responsible AI Officer at the National Security Agency. The group spans CISOs, CEOs, security researchers, and academic faculty across the U.S. and allied countries.

They argue that Mythos-class models can find software flaws and generate exploits, but are not unique in this. The same capabilities exist in GPT-5.5, Opus, Sonnet, and Chinese open models, including Kimi 2.7. The signers contend that removing the best American tools from defenders without restricting what attackers can already access elsewhere weakens defense without limiting offense. The letter also challenges the jailbreak evidence the government cited to justify the directive. The signers say what the jailbreak unlocks in Fable is already freely available in other models without bypassing any safeguards. Fable’s jailbreak, they argue, does not give attackers capabilities they could not already access elsewhere. Anthropic is working to close the specific vulnerability.

If the government does pursue AI risk regulation, the letter says it should be grounded in scientific consensus developed with industry and academic input, created through democratic rule-making, enforced transparently with time to remediate, and applied only as far as necessary to protect public safety.

The letter remains open for additional signatures at freefable.org. The government has not publicly responded.

Butterfly and moth specimens forming the numeral five, cut by a red and white barrier with a stop sign, representing the suspension of Fable 5 and Mythos 5 access.

Anthropic Shuts Down Fable 5 and Mythos 5 After US Government Security Directive

, , , , , , , , , , ,

Unable to verify which customers are foreign nationals, Anthropic cut access for everyone.

 

On June 12, the US government ordered Anthropic to suspend access to Fable 5 and Mythos 5 for foreign nationals, citing national security concerns. To comply, Anthropic shut down access for all users, saying it has no mechanism to verify customer nationality.

The government says it has become aware of a jailbreak for Fable 5. Anthropic disputes the severity, calling it a non-universal jailbreak that works only in limited situations and only enables what other AI models already do freely.

Anthropic complied, but is pushing back. What the government shared, Anthropic says, constitutes a narrow, non-universal technique: one that works only under specific circumstances, not one that broadly defeats the model’s safeguards.

Anthropic said at the time of launch that no AI model can be made completely jailbreak-proof. Their stated strategy was to minimize the damage any jailbreak can cause, make them difficult to execute, and catch them quickly. The 30-day data retention requirement on Mythos-class models is part of that: it lets Anthropic review usage and shut down jailbreaks as they appear.

“If this standard was applied across the industry,” Anthropic said, “we believe it would essentially halt all new model deployments for all frontier model providers.”

 

Background

Mythos 5 and Fable 5 launched three days before the directive. Mythos 5 is available exclusively to vetted large tech companies, US and allied government organizations, and cybersecurity firms through Project Glasswing, Anthropic’s program for organizations that use AI to hunt for software vulnerabilities before attackers find them. 

Fable 5 is the consumer version of the same underlying model, with filters that detect hacking-related requests and route them to a less capable model, Opus 4.8, rather than allowing Fable 5 to respond.

Glasswing launched in April 2026 with Claude Mythos Preview, the predecessor to Mythos 5. Partners in the project identified more than 10,000 vulnerabilities using Mythos Preview, including bugs that conventional security tools had missed.

The relationship between Anthropic and the US government is strained. The US government takes an active role in vetting which organizations can access Mythos 5 through Glasswing. This cooperation exists despite an ongoing lawsuit where Anthropic sued the Department of War after the DoW designated Anthropic a supply chain risk earlier this year.

 

Anthropic says it believes the suspension is a misunderstanding and is working to restore access as soon as possible.

Red dots, black grids and abstract pattern blocks illustrate Anthropic AI policy proposal.

Anthropic Urges Congress To Require testing Of frontier AI Models

, , , , , , , , , , ,

The AI company released a broader policy proposal as lawmakers debate how the United States should govern increasingly capable AI systems.

 

Anthropic is urging Congress to require independent safety testing for the most capable AI models, arguing that existing policy discussions do not go far enough to address the risks posed by rapidly advancing AI.

The recommendation was included in a broader policy proposal released Tuesday, alongside Anthropic’s essay, The Policy on the AI Exponential. The company said the United States should strengthen oversight of frontier AI systems as model capabilities continue to improve and are more widely deployed.

According to Reuters, Anthropic also urged lawmakers not to preempt state AI laws unless a robust federal system is in place and called for independent evaluations of the most advanced models before deployment.

The proposal arrives as Congress continues to debate federal AI legislation and as policymakers weigh how much responsibility should fall on companies developing the technology.

In its essay, Anthropic argued that AI progress is accelerating and could have major consequences for national security, economic growth, and the labor market. The company said governments should prepare for a future where advanced AI systems play a much larger role in society and the economy.

“We may be entering a period of very rapid change,” Anthropic CEO Dario Amodei wrote in the essay, describing the possibility that AI systems could become significantly more capable over the next several years.

Anthropic’s proposal outlines several areas where the company believes policymakers should focus, including AI safety testing, national security safeguards, export controls, infrastructure development, and the government’s capacity to evaluate advanced models.

The company also called for greater transparency around frontier AI development. It argued that policymakers should have access to independent assessments of the most capable systems before they are widely deployed.

The recommendations come as lawmakers, regulators, and technology companies continue to debate whether existing laws are sufficient to address the risks associated with advanced AI models. Several AI developers, including Anthropic, OpenAI, and Google, have publicly supported some form of government oversight for the most powerful systems, although they differ on specific regulatory approaches.

Anthropic’s proposal was released less than a week after House lawmakers unveiled a discussion draft intended to guide future federal AI legislation. The company did not endorse that proposal and instead offered its own recommendations for how Congress should approach AI governance as model capabilities continue to advance.

South lawn view of the White House where President Trump signed NSPM-11 on AI vendor policy in national security, June 2026.

White House Executive Order, NSPM-11, Empowers Agencies to Terminate AI Vendor Contracts Over Usage Limits

, , , , , , , , , ,

The provision formalizes the authority the Department of War used against Anthropic, three months after canceling the company’s $200 million contract.

 

President Trump signed memorandum NSPM-11 on June 5, directing national security agencies to terminate contracts with AI vendors that repeatedly restrict how the government can use their systems. The provision codifies the authority the Department of War used to cancel Anthropic’s $200 million contract in March, after the company declined to permit unrestricted government use of its Claude models.

 

Agencies can terminate AI vendor contracts for imposed restrictions

Section 3(b) of the memo directs agency heads to terminate contracts with any AI company that repeatedly restricts government use of its systems. The memo specifies termination “for default or for convenience,” meaning the government doesn’t need to prove a breach. If an agency determines a vendor’s usage policies conflict with NSPM-11, it can end the contract without establishing fault.

That standard addresses the Anthropic dispute directly. In March, the Department of War terminated its contract with Anthropic after the company declined to accept terms requiring “any lawful use” of its models. Anthropic voiced concerns that its models could be used for mass domestic surveillance and fully autonomous weapons and wanted assurances neither would happen. The Pentagon designated the company a supply chain risk, ordered removal of its products from all defense systems within 180 days, and canceled the $200 million contract.

The rule has an escape valve. Agency heads can keep a restricted vendor relationship active for up to one year if cutting ties would create an immediate national security gap. Each exception must be reported in writing to the National Security Advisor and the President’s top science advisor within 30 days.

 

Agencies now need to use multiple vendors

Section 4(a) directs agencies to diversify their vendor base within 120 days, so they hold active contracts with multiple vendors at once. The directive is a direct response to what happened when the Department of War canceled Anthropic’s contract: agencies had no contracted backup. OpenAI stepped in quickly, but NSPM-11 makes spread-vendor contracting a formal requirement going forward.

 

NSPM-11 rescinds a previous Biden Administration memorandum, NSM-25. The Secretary of War has 90 days to rewrite the Pentagon’s rules on AI in weapons systems. Those rules, set out in DOD Directive 3000.09, govern when and how AI is used in lethal operations, including the conditions under which a human must remain in the decision loop. The revised directive will align those rules with NSPM-11’s requirements and is subject to annual review as AI capabilities advance. A classified annex covering sensitive security matters is expected within the same 90-day window.

The Anthropic case that prompted the memo’s central provision remains in litigation. A California federal court blocked the supply chain risk designation in March, ruling the action was punitive rather than security-based. The D.C. Circuit denied Anthropic’s emergency request to lift the designation in April. The lawsuit continues in both courts.

Hand-drawn illustration of a laptop with a keyhole on screen, representing AI-enabled cyber intrusion and network access risks, 2026.

Anthropic: AI-Enabled Attackers Are Going Deeper, and Harder to Detect

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

AI is now running deeper inside attacks with less human direction. The old tools for spotting the most dangerous actors can’t keep up.

 

Hackers are using AI to penetrate compromised systems more deeply, and chain together attack stages with minimal human input. That combination of deeper access and more autonomy is making the standard signals security teams rely on to identify dangerous actors unreliable. Anthropic published those conclusions in a new report analyzing 832 accounts it banned for malicious cyber activity between March 2025 and March 2026.

 

AI is lowering the bar for getting in

Until recently, writing malware required real programming skill. AI has changed that. In Anthropic’s dataset, 67% of the 832 banned actors used AI to write malware. That matters because attackers who previously lacked the technical ability to build intrusion tools can now get them from a chatbot.

Once inside, AI is taking them further. AI use for account discovery rose 8.9% over the study period. Account discovery is how hackers map out which accounts exist inside a compromised network to find the credentials and access points they need to move deeper. At the same time, AI-assisted phishing, used to gain initial access, fell 8.6%. Attackers are spending less AI effort getting in and more AI effort going deeper once they’re there.

The result: the share of attackers classified as medium risk or higher jumped from 33% in the first six months of the study to 56% in the second.

 

Why it’s now harder to tell the dangerous ones from the rest

Security teams traditionally assess risk by counting how many distinct attack techniques an actor uses and what tools they access. Both signals have broken down.

In the dataset, the least-skilled actors averaged about 16 distinct techniques. The most skilled averaged about 20. Four techniques separate the bottom from the top. The platform used (Claude Code, API, or standard chat) also showed no correlation with an actor’s perceived danger.

What separates the highest-risk actors is whether they have built systems that let AI run the attack automatically, chaining together stages such as account discovery, lateral movement, and privilege escalation with minimal human direction at each step. Less-skilled actors are increasingly doing this too, which explains why the proportion of medium-risk-or-higher attackers nearly doubled in a single year.

That type of automated, multi-stage attack has no classification in the MITRE ATT&CK framework, the standard reference security teams use to track how attackers operate. Anthropic says it is in discussions with MITRE about updating the framework and has added safeguards to its most capable models to detect and block activities such as malware development and mass data exfiltration.

Abstract visualization of a dark cracked sphere with explosive blue and gold rays representing Anthropic's Project Glasswing expansion to critical infrastructure sectors.

Anthropic Expands Project Glasswing to 150 New Organizations Across 15 Countries

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

Power, water, and healthcare companies join the program as Anthropic warns that within a year, AI models will become more powerful, cheaper, and widely available, including to cybercriminals. 

 

Anthropic is extending Project Glasswing to approximately 150 new organizations in more than 15 countries, adding sectors not represented in its initial cohort, including power, water, healthcare, communications, and hardware. Each new partner must meet Anthropic’s security requirements before gaining access to Claude Mythos Preview (“Mythos”).

In April, Anthropic announced that roughly 50 initial partners, including NVIDIA, Apple, Google, Microsoft, and Amazon, had begun using Mythos to scan their codebases. Those partners found more than 10,000 high- or critical-severity security flaws in the first weeks of the program.

The new partners share one characteristic: scale of exposure. For most, Anthropic estimates a successful attack on their codebase could affect more than 100 million people, with consequences for both national and global security. Many are vendors whose software underpins systems relied on by other organizations worldwide, including governments.

Anthropic increased participation out of concern that other AI companies will likely have Mythos-class models within 6 to 12 months and could release them without safeguards that prevent misuse. In that scenario, AI-assisted cyberattacks become significantly more common and harder to predict. The company frames Project Glasswing as a way to establish operating norms before that happens.

Anthropic notes that finding vulnerabilities is no longer the bottleneck: getting them patched is. Partners now use Mythos to write patches and run pre-release checks on their own code, not just to surface problems. Anthropic says the model can also be used for penetration testing, threat detection and response, and rebuilding legacy codebases in memory-safe languages. This week, Google released AI Threat Defense, an advanced cyber detection model that also automatically applies patches.

Anthropic said it plans to further expand Glasswing, prioritizing critical infrastructure providers, open-source software maintainers, and safety testers. It is also scaling its Cyber Verification Program, which would grant Mythos-class capabilities to more organizations for specific defensive tasks. General access to Mythos, the company said, requires safeguards no AI developer has yet built.

Gray Swan co-founders Matt Fredrikson and Zico Kolter, the CEO and Chief Scientist of the AI security firm, following its $40M Series A announcement.

Gray Swan, AI Security Vendor to AI Frontier Developers, Raises $40M Series A

, , , , , , , , , , ,

“Stress test” platform serving the major AI frontier models, including Anthropic, OpenAI, and Meta, secures $40M in Series A funding.

 

The company closed a $40 million Series A co-led by Wing Venture Capital and Madrona. Obvious Ventures, Snowflake Ventures, Hudson River Trading, Samsung Next, and existing investor Magarac Venture Partners also participated.

Gray Swan builds security tools for AI systems. Gray Swan helps enterprises and AI labs find vulnerabilities in AI models and agents before deployment, then block dangerous behavior in real time. Its platform has three components: Cygnal, which monitors AI outputs and enforces custom security policies set by the organization; Shade, which automatically runs adversarial tests against AI deployments before they go live; and Arena, a global competition where more than 15,000 security researchers probe AI models for weaknesses. Arena has generated more than 1 million attack trajectories that feed the threat intelligence for Cygnal and Shade.

Gray Swan’s adversarial evaluations are cited in 11 recent frontier model system cards, including those from Anthropic, OpenAI, and Meta. System cards are the safety documents labs publish before releasing a model to the public; Gray Swan’s testing is embedded in that process. The company says no other AI security vendor holds that position across all three major labs simultaneously. It works with more than 20 customers across frontier labs and global enterprises.

 

Snowflake partnership

Gray Swan also announced a partnership with Snowflake. Its real-time protection now runs inside Snowflake’s AI platform. Enterprises that build and run AI applications on Snowflake can apply Gray Swan’s security layer directly in that environment, without a separate integration step. Snowflake is one of the primary platforms where enterprises deploy AI on their own data, making the integration relevant to companies already in that stack.

 

The funding will go toward sales expansion, deeper lab partnerships, and team growth.

 

Load More