Latest Models Upend White House AI Plans

 

If some technology, like Mythos, is important for safety of all local governments in America, federal government must pay to apply it. Talk is cheap. Regulation is misguided. You cannot regulate all the world. You cannot regulate even Iran.

 

As of May 2026, the rapid advancement of powerful AI models like Anthropic's Mythos is causing a significant shift in White House AI policy, with officials considering strict oversight for models capable of finding and exploiting software vulnerabilities.

•           White House AI Reset: Driven by alarms from Vice President JD Vance and others regarding Mythos's ability to disrupt local infrastructure (water plants, hospitals), the administration is weighing an executive order that could treat AI models similar to FDA-regulated drugs—requiring safety vetting before release.

•           Security Concerns: Anthropic has temporarily limited access to Mythos—offering it only to select partners—due to its high-risk, high-reward ability to identify software flaws at an unprecedented scale.

•           Federal Action on Local Safety: While the federal government (via DHS/CISA) is being urged to secure critical local systems, current policy focuses on a mix of partnership and potential regulation. The administration is also leveraging existing grant programs to help local governments adopt secure AI tools.

•           Voluntary Cooperation: Major companies like Google and Microsoft have agreed to allow federal review of new AI models for security risks.

How Local Governments Are Accessing AI Safety Tools

The federal government is not necessarily paying for, or applying, specific private AI models like Mythos directly at the local level yet. However, the government is focusing on:

•           Funding and Guidance: Existing initiatives like the State and Local Cybersecurity Grant Program (SLCGP) at CISA provide nearly $1 billion to help local governments secure their systems.

•           Proactive Defense: The White House's National Policy Framework for Artificial Intelligence (March 2026) aims to foster secure, "industry-led" standards.

•           AI-Enabled Grants: New AI-powered tools are being used by local governments to navigate the complexities of federal grant applications to pay for their own technological upgrades.

 

 

 

“WASHINGTON -- On a recent call with the heads of the biggest artificial-intelligence companies, Vice President JD Vance was alarmed.

 

New AI models such as Anthropic's Mythos, which are capable of finding software vulnerabilities on their own, threatened to disrupt small town banks, hospitals and water plants by launching cyberattacks that local governments weren't equipped to handle, Vance said.

 

"We all need to work together on this," Vance told the chief executives, among them OpenAI's Sam Altman, Dario Amodei of Anthropic, Elon Musk of SpaceX, Sundar Pichai of Alphabet's Google and Microsoft's Satya Nadella, according to people familiar with the matter.

 

The April call, which followed a White House briefing that played a role in sparking Vance's worries over the latest AI model capabilities, set in motion a chaotic administration response to Mythos that threatens to increase government oversight of AI and overhaul the administration's tech agenda. The concern expressed by Vance, paired with other moves by the White House to get involved in the rollout of AI models, mark a shift from previous language about winning the AI race against China and removing barriers to deploying models.

 

The White House is now weighing an executive order that could create a formal oversight process for the most advanced models. Administration officials have also asked Anthropic to hold off on expanding access to Mythos to more companies and organizations managing critical digital infrastructure. The White House has tapped National Cyber Director Sean Cairncross to lead its response to the model, which could include other actions to address model safety or limit the ability of private companies to dictate how the government uses their tools.

 

The White House concern over Mythos has left some administration officials and congressional aides fearful that it represents a reversal on AI policy and an overreaction, people familiar with the dynamics said. Meanwhile, moves by Cairncross to control access to Mythos and the government's response have frustrated administration officials who want more say in the process, those people said.

 

The new executive order under consideration has been cheered by proponents of AI safety as a potential rebuttal of the hands-off approach led by White House adviser David Sacks, a venture capitalist. "People are treating this like some existential threat. I don't think it is as long as everyone does what they're supposed to do" by using the AI tools to bolster digital security, Sacks said last week on the "All-In" podcast he co-hosts.

 

Administration officials said they are balancing proper oversight and innovation. "The White House will continue to lead an America First effort that empowers America's great innovators, not bureaucracy, to drive safe deployment of powerful technologies," White House chief of staff Susie Wiles said Wednesday on X.

 

The White House remains committed to an industry-friendly AI strategy while aiming to address the risks posed by Mythos and other powerful models such as OpenAI's GPT-5.5-Cyber, an administration official said. Any discussion about executive orders is speculation unless announced by President Trump, a White House official said.

 

OpenAI consulted the administration before recently previewing its most advanced cyber model, which has demonstrated capabilities similar to those shown by Mythos, a spokeswoman said. OpenAI is also limiting access.

 

Washington and Beijing are weighing the launch of official discussions about the risks posed by AI ahead of next week's summit with Trump and China's Xi Jinping.

 

White House National Economic Council Director Kevin Hassett, speaking Wednesday on Fox Business, compared the expected oversight process for advanced AI models to the process drugs go through before they are cleared by the Food and Drug Administration. The comments fueled criticism from allies of the administration that the White House was considering burdensome rules that would slow down AI companies.

 

"Importing the FDA approach into AI would upend President Trump's current pro-growth AI policy," said Neil Chilson, head of AI policy at the Abundance Institute, a nonprofit that opposes excessive government intervention in the private sector.

 

The vice president has taken a role as an intermediary between tech companies and the administration, and the April call was a rare moment for top administration officials to speak to so many tech CEOs at once.

 

Vance helped negotiate a deal to give investors friendly with the White House control of TikTok's U.S. operations. When Sacks and other White House advisers were pushing Congress to pass a moratorium on state laws regulating AI late last year, Vance served as a go-between for proponents of the ban and MAGA allies of the White House who were fighting it. Trump signed an executive order in December targeting state rules but has promised to protect children online and limit the damage of AI data centers.

 

Treasury Secretary Scott Bessent, as well as Secretary of State Marco Rubio, joined Vance and Cairncross on the April call. The Treasury secretary has warned top bank executives about the risks posed by Mythos, which is so good at finding software vulnerabilities that Anthropic isn't releasing it publicly.

 

Groups arguing for more oversight have cheered the administration's recent comments after opposing the White House's moves to dismantle the Biden administration's tech strategy, which included plans for reviewing models before release.

 

"This is a promising step in the right direction and a fundamental shift in how the federal government is thinking about its role," said Brad Carson, president of Americans for Responsible Innovation, a bipartisan nonprofit that pushes for stronger guardrails.” [1]

 

1. Latest Models Upend White House AI Plans. Ramkumar, Amrith; Schwartz, Brian; Andrews, Natalie.  Wall Street Journal, Eastern edition; New York, N.Y.. 08 May 2026: A1.