Asking a model to fix your code is now a national security jailbreak

A Friday-evening letter that bans a whole class of users

Anthropic's public statement opens with a timestamp: the directive arrived on June 12, 2026, at 5:21pm ET. "The US government, citing national security authorities, has issued an export control directive to suspend all access to Fable 5 and Mythos 5 by any foreign national." Every other Anthropic model stays online; only the two flagships go dark for non-citizens, and in practice for nearly everyone.

The striking part is what the government actually showed. By Anthropic's account the only evidence was verbal, and the concern is "a potential narrow, non-universal jailbreak, which essentially consists of asking the model to read a specific codebase and fix any software flaws." Anthropic says it reviewed the report and found the capability "widely available from other models (including OpenAI's GPT-5.5), and is used every day by the defenders who keep systems safe." After thousands of hours of red-teaming with the government and UK AISI, it adds, it has "not even received a disclosure of a concerning non-universal potential jailbreak that led to a harmful result."

The company's own line on principle is doing heavy lifting in the thread. Anthropic restates that it believes the government should be able to block unsafe deployments "as part of a statutory process that is transparent, fair, clear, and grounded in technical facts," then says this action "does not adhere to those principles" and that applying this standard would "essentially halt all new model deployments for all frontier model providers."

"The 'jailbreak' is to ask it to fix your codebase"

The technical detail in the thread drew the sharpest reaction, because the banned capability is the product's whole point.

"If I read that right, the 'jailbreak' is to ask the model to fix the codebase and then it exposes the flaws? That sounds like a gap that is nearly impossible to fix while retaining high capability. Like you want it to be able to fix your codebase…" — an engineer parsing the disclosed technique

One reader pulled Anthropic's own paragraph forward to argue the trigger had nothing to do with the company's safety messaging:

"the level of capability displayed there is widely available from other models (including OpenAI's GPT-5.5), and is used every day by the defenders who keep systems safe." — a commenter quoting the statement to show the parity with GPT-5.5

Four camps, one thread

The first camp reads this as karma for years of capability hype.

"Finally they will pay for all the scaremongering they been doing to sell their models as something so much ahead of all else." — a software engineer in the thread

"Well maybe now they will learn that they shouldn't overhype the capabilities of their models." — a developer reacting to the suspension

The second camp reads the same statement and sees Anthropic getting exactly the regime it lobbied for.

"This is what Anthropic wanted and they want this to apply to all other frontier models providers (including themselves) that release powerful models… They ultimately got what they wanted." — a reader quoting Anthropic's statutory-process line back at it

A third camp reads it as a personal score being settled.

"I would more easily guess that it is a revenge of Trump for Anthropic humiliating him when he wanted to use it without control for military purpose. And indeed it used against them their own marketing allegations." — a longtime HN regular

And a fourth camp, the cynics, sees a marketing gift dropped right before an IPO.

"Sadly, I suspect this will be the best piece of marketing they could ever hope for. 'It's so advanced the government made us add extra security* to stop hackers!'" — one skeptical reader

The credentialed counter: this is bigger than Anthropic

Not everyone treats it as drama. A neuroimaging scientist who works in the AI space pushed against the mockery:

"You call it scare-mongering. Others, serious thinkers and leaders in the AI and national security space, believe, maybe not scare-mongering enough. AI is a national security issue. Best accept that as fact, or you won't see it coming." — a neuroimaging scientist

Others warned the schadenfreude misses who actually loses.

"This affects more than just Anthropic. It's a game of thrones and everyone using this technology loses. I wouldn't cheer for that." — another voice in the thread

One argued the whole monetization story for ever-larger models just took a hit:

"there goes the semiconductor boom predicated on monetization of ever larger models… Once the US government acts out of capricious fiat because a model becomes 'too good' and they demonetize it, the entire shell game collapses." — a longtime HN user

And the darkest read questioned whether you can trust an Anthropic model on your codebase at all now:

"Likely models by Anthropic can no longer be reliably trusted as it'll subtly sabotage your codebase you're working on. Gov just need some national security orders for Anthropic not disclose it to the public…" — a more paranoid reader

Why this one is a rubicon

Strip the politics and a precedent remains: a government can, on verbal evidence, switch off the strongest commercial model overnight for an entire class of users. Whether the motive was safety, revenge, or favoritism, the mechanism now exists and has been used once. The proposed call to action in the thread — that allies should nationalize weights already deployed in their jurisdictions — went nowhere on feasibility, but the underlying worry did not. If you cannot run the model on hardware you own, it can be taken away on a Friday afternoon, and you will read about it in a timestamp.