BitGo CEO Disputes Claims of Anthropic Mythos NSA Breach

BitGo CEO Rejects Claims That Anthropic’s Mythos Hacked NSA Systems – Clarifies Test Context Amid Viral Reports

Key Takeaways

• BitGo CEO Mike Belshe publicly rejected claims that Anthropic’s Mythos model breached NSA classified systems.
• The claim originated from remarks by Senator Mark Warner, as reported by The Economist.
• The incident described was an authorized red-team test on NSA systems, not an external intrusion.
• Anthropic disabled the model on June 12 due to a US export-control directive, not because of a security breach.

How the Claim About an NSA Breach Emerged

The controversy began after Senator Mark Warner, vice chair of the Senate Intelligence Committee, described a briefing from General Joshua Rudd, who leads the NSA and US Cyber Command. According to The Economist, Warner recounted that Rudd said Anthropic’s Mythos model “broke into almost all of our classified systems, not in weeks, but in hours.”

Warner cited the example while expressing support for Anthropic. He used the case to argue for faster pre-release testing of advanced artificial intelligence models. However, as the quote circulated on social media platform X, it was increasingly presented as evidence of a real-world cyberattack on US intelligence systems.

Shashank Joshi, the Economist editor who published the quote, later clarified that the statement should not be interpreted literally. He noted that the scenario depended on Mythos operating alongside other tools under specific conditions. He also emphasized that the episode described an authorized test, not an unauthorized breach.

Authorized Red-Team Test, Not External Intrusion

The central detail missing from many online posts is that the activity in question took place as part of a sanctioned red-team exercise on the NSA’s own networks. Red-team tests are designed to simulate attacks in order to identify vulnerabilities. They are conducted with prior authorization and oversight.

At the time of the test, the US government was already working with Anthropic. Since April, the company had deployed Mythos to government cyber defenders through Project Glasswing. The testing therefore occurred within an established partnership framework.

Several analysts and commentators highlighted this distinction. Zack Korman pointed to how the claim evolved from a senator’s remarks to a journalist’s report and then into viral social media posts. Analyst Kyle Chase stated that the system access occurred during a controlled test and not as a hostile intrusion. He also referenced a separate jailbreak issue flagged by Amazon as the more direct trigger for subsequent action.

BitGo CEO Publicly Challenges the Narrative

Mike Belshe, co-founder and chief executive of digital-asset custodian BitGo, responded directly to one of the social media threads amplifying the breach narrative. He wrote that he was “calling BS on this,” disputing the framing that suggested a successful hack of classified systems.

Belshe’s intervention brought attention from the cryptocurrency sector, where infrastructure providers such as custodians closely monitor cybersecurity developments. While his statement did not introduce new technical details, it underscored skepticism among industry executives regarding how the episode was portrayed online.

For users of crypto platforms and digital asset services, cybersecurity claims involving US intelligence agencies can influence perceptions of systemic risk. In this case, the dispute centers on whether the event represented a real compromise of classified systems or a controlled security evaluation.

Model Shutdown Linked to Export Controls

Anthropic stated that it disabled Mythos and another model on June 12 in order to comply with a US export-control directive. According to the company, the decision was not driven by any operational breach of government systems.

The company explained that the jailbreak in question involved prompting the model to review a codebase and identify flaws. This process revealed several minor, previously known bugs. Anthropic stated that comparable models, including OpenAI’s GPT-5.5, are capable of identifying similar issues.

Anthropic objected to the recall of a model used by hundreds of millions of people over what it described as a narrow flaw. The company warned that withdrawing models for isolated vulnerabilities could halt the release of new systems across the sector. It is currently working to restore access and is drafting a shared risk framework with the White House.

AI researcher Pedro Domingos argued that export controls were the decisive factor behind the shutdown, given the model’s strong cyber capabilities. Anthropic itself has described Mythos as the strongest cyber model in the world.

Why the Distinction Matters for Digital Infrastructure Users

For participants in cryptocurrency markets and digital services, clarity around cybersecurity incidents is essential. Reports suggesting that an AI model penetrated classified US systems can affect trust in both AI providers and the broader technology ecosystem.

In this case, the available information indicates that the system access occurred during an authorized internal test and within a formal partnership between Anthropic and the US government. The shutdown of the model followed an export-control directive rather than confirmation of an uncontrolled breach.

For users evaluating technology partners, infrastructure providers, or platforms that rely on advanced AI systems, the difference between a sanctioned stress test and a hostile cyberattack is significant. The episode demonstrates how rapidly statements can shift in meaning as they move from official briefings to media coverage and then to social networks.

Our Assessment

Based on the reported facts, the claim that Anthropic’s Mythos model breached NSA classified systems originated from a description of an authorized red-team test. Subsequent clarifications from the reporting journalist and statements from Anthropic indicate that no external intrusion occurred. The model’s shutdown was linked to a US export-control directive rather than a confirmed security compromise. Public responses from figures such as BitGo CEO Mike Belshe reflect concerns about how the incident was framed as it circulated online.