Welcome to the latest edition of the AI Update, where we continue to track the major developments in artificial intelligence and their implications for identity security and cybersecurity more broadly.
This week, Anthropic continues to be the main character after the announcement of its Mythos Preview model last week. Despite its ongoing legal dispute with the Pentagon after being labelled a “supply chain risk”, Anthropic is now in talks with the White House about the potential use of its models by government agencies. The White House Office of Management and Budget is reportedly preparing guardrails that could let major federal agencies use Mythos – which is arguably a powerful cyberweapon – though it looks like it could end up being a modified version of the model that gets approved. As Anthropic co-founder Jack Clark explained, “We have a narrow contracting dispute, but I don’t want that to get in the way of the fact that we care deeply about national security.”
Not to be totally left in the dust, OpenAI has launched its own cybersecurity-focused model, along with a limited-access program for partners to test it. OpenAI says that GPT-5.4-Cyber is a specialized version of its GPT-5.4 model that is specifically designed to search for cybersecurity vulnerabilities. OpenAI is letting partners access the model via its Trusted Access for Cyber program, “an identity and trust-based framework designed to help ensure enhanced cyber capabilities are being placed in the right hands,” and is offering a $10 million API-credit commitment for cyber defense.
Meanwhile, Google is in talks to let the U.S. Department of War use its Gemini AI models in classified environments. Like Anthropic before it, Google is reportedly looking to get assurances that its technology won’t be used for domestic surveillance or in autonomous weapons systems, while the Pentagon is insisting on being able to leverage the AI tech for “all lawful uses.” A Pentagon spokesperson said the military “will continue to rapidly deploy frontier AI capabilities to the warfighter through strong industry partnerships across all classification levels.”
Cerebras is reportedly preparing to go public in the second quarter of this year. The California-based company specializes in AI chips, and is riding the high of a recently expanded partnership with OpenAI that will see the latter spend more than $20 billion over the next three years to use servers powered by Cerebras chips. The company had previously withdrawn an IPO in October of 2025 amid a national security review tied to a past G42 investment. It later obtained CFIUS clearance, demonstrating that it doesn’t pose any national security risk from the US government’s perspective.
A new survey from the firm Grant Thornton found that 78 percent of the 950 senior US business leaders polled said they were not confident that their organization was using AI safely – specifically, responding that they weren’t sure their organization could pass an independent AI governance audit within 90 days. Only 12 percent said their organization is AI-ready. At the same time, almost three-quarters of respondents indicated that their organizations were already running autonomous or semi-autonomous systems internally.
Circling back to Anthropic, the startup has quietly added a selfie-based identity verification process to its platform, tying it to access to certain Claude capabilities. The process has been implemented in partnership with Persona, and involves matching a live selfie to images of government-issued photo ID. But it’s not yet clear how this identity verification system will be implemented for end users, with Anthropic’s help desk page only vaguely referencing “certain capabilities” and “routine platform integrity checks.”
A humanoid robot went viral this week when it was filmed chasing a group of wild boars off the streets of Warsaw. Dubbed “Edward Warchocki”, the robot appears to be a customized version of the Unitree G1 model, and is ostensibly tackling a real problem in Warsaw, where boars have been wandering in recent years and occasionally attacking dogs. In the video, Edward shouts “Go away!” in Polish, and seems to shake his fist at the boars as they escape into the distance.
And we conclude this week with a delightful meme about Claude Mythos’ exceptional red-teaming capabilities, care of Alex Krusz on X (@AlexKrusz), and shared in the latest Control Plane newsletter:
Thanks for reading!
–
By Alex Perala
