BitGo CEO Mike Belshe has rejected a viral claim that Anthropic's Mythos AI breached nearly all NSA classified systems, days before Washington cleared the model for more than 100 partners.
Key Points
- A viral post said Mythos cracked nearly all NSA classified systems in hours, though the event was an authorized red-team test.
- Belshe and several analysts called the hack framing false, noting that no official NSA statement ever confirmed it.
- Washington pulled Mythos under a Jun. 12 export order, then cleared it for more than 100 trusted partners on Jun. 26.
Mythos NSA Claim Traces Back To Senator
The breach story spread fast across X this weekend, recast as a live hack of the nation's most guarded secrets, and the name Mythos quickly trended. The fuller record is narrower. Senator Mark Warner, vice chair of the Senate Intelligence Committee, first relayed the account from a closed briefing that drew little notice until an AI commentator amplified it over the weekend.
He said NSA director Gen. Joshua Rudd had described the model cracking almost all classified systems in hours, not weeks, during an authorized Jun. 11 exercise on the agency's own networks. Warner offered the case to praise Anthropic and to press for mandatory pre-release testing of frontier models, and the writer who first published the line later clarified that it should not be read literally.
Also Read: Chainlink's Wallet Record Turns LINK's $9 Rebound Into The Main Test
Belshe, Analysts Reject Hack Framing
Belshe, who co-founded the digital-asset custodian BitGo, dismissed the viral version outright.
He called the framing false in a widely shared reply. Several analysts piled on, since no official NSA statement ever backed the central claim, and a senator's offhand line had carried it online.
Analyst Kyle Chase noted the break-in was a controlled drill, the kind security teams run to expose weak points before adversaries do. He said the real trigger was a separate jailbreak that Amazon had flagged to the government weeks earlier, not the NSA exercise.
That technique simply asked the model to read a codebase and fix flaws. It turned up a few minor, already-known bugs that rivals like OpenAI's GPT-5.5 can also find. A routine government test had warped on its trip from a closed hearing to viral posts over a single weekend.
Mythos Freeze Eases After Two-Week Standoff
Anthropic disabled Mythos and its weaker sibling Fable 5 on Jun. 12 to meet a US export-control order, not over any battlefield breach, the company has said. The model had quietly served federal cyber defenders since April through Project Glasswing, a closed program covering roughly 200 vetted firms.
Dozens of cybersecurity experts had urged the White House to lift the curbs, calling the models capable but not unique. On Jun. 26, the government allowed Mythos back for a vetted set of cyber defenders and infrastructure providers, more than 100 institutions in all. Fable 5 stays offline, and the company is now drafting a shared risk framework with officials to govern future releases.
Read Next: XRP Falls Near $1 While ETF Buyers Test A Weak Spot Market