How powerful is Anthropic’s Claude Mythos? Powerful enough to be a national security threat, as Anthropic itself warned? Powerful enough to create better clones of existing video games and vibe coding services? A new revelation has revealed the extent of Mythos’ capabilities – the AI model is powerful enough to break into the US government’s classified networks in just hours. Not weeks, but hours.
Senator Mark Warner (D-VA), vice chairman of the Senate Intelligence Committee, disclosed that the Mythos model developed by Anthropic was able to penetrate “almost all” of the US National Security Agency’s classified systems in a matter of hours rather than weeks.
The claim, attributed to General Joshua Rudd, director of the NSA and US Cyber Command, was shared during a June 11 briefing and later reported by The Economist. The extent of the compromise remains unclear. No public information has been released about specific systems affected or the precise methods Mythos employed.
Anthropic’s Mythos claimed to be mighty powerful
Anthropic’s Mythos class models have been advertised as mighty capable AI models that organisations around the world are relying on to secure their infrastructure. In this case, several cyber experts have commented that this hack attempt was not an unauthorised attack but a controlled exercise that deliberately wanted to uncover weaknesses.
Mythos reportedly surfaced thousands of high-severity vulnerabilities and demonstrated autonomous exploit chaining with minimal human oversight. One of the most talked-about examples was a rapid exploitation of a 17-year-old remote code execution flaw in FreeBSD following a single prompt. This only highlights how modern AI can accelerate what once required teams of skilled human hackers working over extended periods.
The revelation comes at a time when the entire world is concerned over the Mythos controversy. US has gotten busy ever since the Anthropic deal with the Pentagon turned sour, with the recent episodes including a ban on the export of the Mythos-class models to other countries. This led Anthropic to eventually pull the plug on Claude Fable 5 (a restricted Mythos), since there’s no way for the AI firm to verify the nationality of an individual before serving its most advanced models.
Will Anthropic’s Fable 5 return to the masses?
There seems to be a possibility, albeit with strict guardrails.
Anthropic has strongly pushed back against a permanent recall of its Mythos-class models. The company argues that completely blocking its most advanced tools over minor compliance issues would hamper Western AI innovation and leave critical infrastructure more vulnerable to foreign adversaries who face no such regulatory friction.
Currently, Anthropic is actively collaborating with the White House and federal regulators to build a joint risk-management framework prior to re-releasing the Mythos-class models.
On the other hand, world governments have encouraged the focus on sovereign AI and technology, urging local players to put all their efforts in building their own AI infrastructure, all in a bid to bypass geopolitics while accessing advanced technologies for their own good.