FTC Disclosure: TechSifted uses affiliate links. We may earn a commission if you click and buy — at no extra cost to you. Our editorial opinions are our own.
The White House has officially told Anthropic: you can't give more people access to Mythos.
That's the headline from a Wall Street Journal report that broke Wednesday evening, confirmed by Bloomberg and a string of follow-on outlets. Administration officials informed Anthropic they oppose the company's plan to expand access to its controversial Mythos AI model from roughly 40 organizations to around 70 — and possibly up to 120 if you include the government agencies being considered on a separate track.
The reasons are specific. And the context — which involves a Pentagon blacklisting, a court fight, a presidential about-face, and a security breach that happened on Mythos's literal launch day — is a lot more complicated than any single news headline can hold.
Let me walk through it.
What Mythos Actually Is
If you haven't been following closely, a quick catch-up.
Mythos is an AI model in a category we haven't really seen before: a system purpose-built for elite offensive and defensive cybersecurity work. Not "help me write a phishing email" basic stuff. Mythos can autonomously find and chain zero-day vulnerabilities in real production systems — software it's never been trained on — and complete multi-step attack sequences from start to finish, without human guidance.
In testing before launch, it autonomously discovered thousands of zero-day vulnerabilities across major operating systems and browsers. More than 99% of those vulnerabilities were still unpatched when Anthropic made its announcement. It succeeded on 73% of expert-level capture-the-flag tasks. And it became the first AI model to complete a 32-step simulated corporate network attack end-to-end — the kind of attack sequence that would ordinarily require a skilled human red team working for days.
Anthropic decided this model couldn't be released publicly. Too much capability, too few guardrails that could realistically contain it at scale. Instead, on April 7, they announced Project Glasswing — a restricted consortium capped at roughly 40 organizations, including Amazon, Apple, Google, Microsoft, NVIDIA, and JPMorgan. Partners received up to $100 million in usage credits, but only for defensive security work. Offensive use: explicitly prohibited under the terms.
The argument for restricted release rather than no release was straightforward: the same capability that could enable mass attacks could also enable organizations to find and patch vulnerabilities before attackers do. Give it to serious defenders, keep it away from everyone else.
We covered that story in depth when it broke: Anthropic Built an AI So Good at Hacking That It Won't Release It to the Public.
The Expansion Plan That Triggered the Pushback
Anthropic had been planning to grow Project Glasswing — from the original ~40 organizations to roughly 70, with potential expansion further if federal agency onboarding proceeded through a separate government track.
That's where the White House drew its line.
Administration officials cited two distinct concerns, and they're worth taking seriously on their own terms.
The first is about security and misuse. Mythos is dangerous enough that even a controlled expansion carries real risk. The same day Anthropic announced Project Glasswing in April, a small group of unauthorized users in a private online forum managed to gain access to the model — the kind of access-control failure that should make everyone nervous. If an AI system capable of autonomous zero-day discovery is hard to contain at 40 organizations, it's harder at 70, and harder still at 120. Each additional node in the access graph is another potential leak point.
The second concern is more operational: compute capacity. Anthropic doesn't have unlimited infrastructure, and administration officials worried that expanding Mythos to 120 entities would degrade the government's own ability to use the model reliably. The concern isn't just that bad things might happen — it's that Anthropic might not have the resources to handle the demand while maintaining service quality for existing users, including US government agencies.
Two different objections. Both credible.
The Backstory Nobody's Quite Connecting
Here's the part that makes this story genuinely strange — because the same administration opposing Mythos expansion spent the previous two months trying to ban Anthropic from the federal government entirely.
In late February 2026, President Trump directed all federal agencies to stop using Anthropic's AI technology. Defense Secretary Hegseth designated Anthropic a "supply chain risk" — a designation normally reserved for companies with ties to adversarial foreign governments. The underlying dispute: Anthropic had refused to allow the Pentagon to use Claude for domestic surveillance and fully autonomous weapons. The DOD wanted unrestricted access across all "lawful purposes." Anthropic said no.
A federal judge blocked the Pentagon's designation in late March, calling it "Orwellian" and writing that "nothing in the governing statute supports the notion that an American company may be branded a potential adversary and saboteur of the U.S. for expressing disagreement with the government." Strong language. The appeals court declined to immediately reinstate it in early April.
And then — while that legal battle was still unfolding — Anthropic briefed Trump administration officials on Mythos capabilities directly. TechCrunch confirmed that briefing happened on April 14. By April 21, Trump was publicly saying Anthropic was "shaping up" and a deal was "possible" for Department of Defense use.
So within roughly 60 days: Anthropic went from supply chain risk banned from the federal government to "shaping up" with a potential Pentagon deal on the table. That's a remarkable pivot, and it was clearly enabled by the Mythos briefing.
Which makes the White House's current opposition to the expansion plan read differently. This isn't "we don't trust Anthropic." It's closer to "we want to control who gets access to this capability and under what conditions." That's more sophisticated — and arguably more legitimate — than the blunt-instrument supply chain designation that a federal judge threw out.
There's a negotiation happening here, and the Mythos expansion question is a bargaining chip. Whether you find that reassuring or troubling probably depends on how much faith you have in the administration's judgment about AI risk.
What This Means for Enterprise Claude Users
If you're running Claude in your enterprise stack right now, the Mythos story probably feels distant from your actual work. You're using Claude Sonnet 4.6, maybe Opus, for coding, analysis, document processing, internal tools. The Mythos drama is in a different category.
That's mostly right. Nothing about this story changes what's available to standard Claude enterprise customers. The models you can access today, you can still access. The APIs haven't changed. Our full Claude AI review reflects capabilities that remain unaffected by this fight.
But there are two things worth watching if you're planning AI procurement.
Access limbo for the ~70. If your organization was in Anthropic's queue for expanded Project Glasswing access — expecting to be part of the expansion from 40 to 70 — you're now in an indefinite holding pattern. That's a real disruption for security teams that had built roadmaps around Mythos availability. The administration's opposition doesn't have a clear resolution timeline, and Anthropic's public response so far has been quiet.
The regulatory signal. The more important implication isn't Mythos specifically — it's what this fight reveals about where US AI policy is heading. The government is now asserting an active role in deciding who can access advanced AI capabilities, even from private US companies. Not through legislation. Not through formal regulatory process. Through informal pressure, procurement restrictions, and supply chain designations.
I've spent the last few months talking to enterprise teams about AI vendor evaluation, and this is a new variable in those conversations. The question used to be: does this model work for my use case? Now there's a second question: what's the regulatory and political status of the company behind this model, and how might that change over the next 18 months?
That uncertainty has a cost. It's not hypothetical anymore.
The Bigger Picture: How US Government AI Oversight Is Actually Taking Shape
What the White House is doing here isn't unique to Anthropic — it's part of a broader pattern of the federal government asserting influence over AI development through non-legislative mechanisms.
Export controls on AI chips. Procurement restrictions. Executive orders covering federal AI use. Defense Department supply chain designations. And now informal pressure on private AI companies about who can access their most capable models. Each of these is a different tool, and none of them required new legislation.
That's genuinely novel. Previous major government interventions in US tech markets — antitrust actions, export controls, national security reviews through CFIUS — had established legal frameworks with defined processes and appeal rights. What's happening with Anthropic and Mythos is messier. The supply chain designation was legally dubious enough to get blocked within weeks. The current White House opposition isn't even a formal legal order — it's pressure.
Pressure works without legal authority when the company cares about its government relationships. Anthropic clearly does — hence the Mythos briefing, the "shaping up" signal from Trump, the White House guidance draft about federal agency onboarding. There's an active negotiation, and the administration has leverage precisely because Anthropic wants a path back to federal contracts.
If the emerging precedent is that the government can informally shape AI capability distribution through pressure campaigns, that's a fundamentally different regulatory environment than anything the tech industry has navigated before. The cloud era, the mobile era, the social media era — all developed under a regime where the government mostly reacted after the fact. This is something different: proactive, informal, and operating faster than any legislative process.
Whether that's the right approach to AI risk is a genuinely hard question. The capabilities in Mythos are real and serious. The security concerns the administration raised aren't invented. But "informal White House pressure" is a fragile governance mechanism that depends heavily on who's in the room and what they want — which can change in an election.
What to Actually Watch For
A few threads worth tracking over the coming weeks.
Does Anthropic expand anyway? The White House opposition isn't a legal injunction. Anthropic could proceed with the ~70 organization plan and accept the political consequences. Their response to the WSJ report has been conspicuously quiet. That silence is probably strategic, but it won't stay strategic forever.
The federal agency onboarding track. While opposing private expansion, the White House is simultaneously drafting guidance to let federal agencies onboard Anthropic models, including Mythos. Those two things can't coexist indefinitely. Either there's a broader deal that covers both tracks — private and federal — or the tension resolves in Anthropic's favor, the government's favor, or just... doesn't resolve.
The Pentagon litigation. The supply chain designation is still technically alive despite the court injunction blocking enforcement. The underlying legal fight continues, and its outcome shapes Anthropic's long-term negotiating position.
And one note for anyone evaluating Claude for enterprise use: do that evaluation on product merit, not political story. The Mythos situation is real, but it's happening in a different layer of Anthropic's business than the products available to enterprise customers. What matters for your deployment is capability, reliability, and cost — and on those dimensions, the picture hasn't changed.
The Mythos story isn't finished. It's just entered its most consequential phase.
Sources: Bloomberg / WSJ · Axios · Washington Post · CNBC — Trump on Anthropic · TechCrunch — Mythos briefing · CNN — Pentagon injunction · AFP / France24
Top comments (0)