How a scary AI-hacking claim traveled general to senator to print

A quote that outran its own headline

The submitted Hacker News title reads like a verdict: an AI model "broke into almost all of our classified systems, not in weeks, but in hours." It is a pulled quote, not the thesis of the piece it comes from. The actual Economist briefing is headlined "Donald Trump's blocking of Anthropic is capricious and chaotic," and the scary line is deployed as evidence inside that argument.

Here is the sentence as it actually appears: on June 11th, Mark Warner, vice-chair of the Senate Intelligence Committee, said that General Joshua Rudd, who leads the NSA and the Pentagon's Cyber Command, had told him that Mythos "broke into almost all of our classified systems, not in weeks, but in hours." Read it slowly. The claim passes from a general, to a senator, to a magazine, before it reaches you. The most quotable AI-security statement of the week is third-hand.

Three relays, none of them an engineer

The sharpest response in the thread is about provenance, not capability. One commenter laid out the chain of custody.

"This quote from TFA is highly likely to be a conflation, exaggeration or extrapolation of what actually happened… It's a paraphrase of a 2nd hand conversation and (at least) the last two 'telephone game' recipients are a U.S. Senator and a general, not security domain or IT experts." — a commenter dissecting the source's chain of attribution

Then the people with actual access to those networks weighed in, and they were unmoved — for the opposite reason. The systems were never airtight to begin with.

"I got in a LOT of trouble for identifying and outlining a trivial privilege escalation attack that worked on both NIPR and SIPR… to my knowledge they never patched it." — a commenter with first-hand experience on classified DoD networks

If the holes are old and known, then a model finding them is not a breakthrough. It is inventory.

"Mythos didn't find new ways to break security, it just went down the list of common security exploits and exposed them for being common even among government agencies." — a commenter characterizing the mechanism as commodity exploitation

That deflates the superhacker framing from one side. Another commenter deflated the Anthropic-specific framing from the other.

"If mythos can break into almost all of their classified systems in hours then other models including opus, gpt, gemini and large open weight models can do so as well… there is no 'maybe' in here." — maintainer of preludejs

If the capability is general, then singling out one lab tells you about the politics, not the threat.

Why the provenance is the story

The non-partisan point survives whatever you think of the policy: a claim about classified systems was generated by two people who do not patch them and printed to support a decision the briefing calls capricious. None of that means the breach didn't happen. It means the most-quoted sentence about it has been through a telephone game whose last two players were chosen for their seniority, not their packet captures. When a number — "hours" — does that much rhetorical work, the cheapest thing to check is who counted.