Senator Mark Warner revealed on 11 June that General Joshua Rudd, who heads both the NSA and Cyber Command, had told him directly: Mythos “had breached almost all our classified systems, not in weeks, but in hours.” The information, reported by The Economist, has not been officially confirmed by the agencies concerned.
The test was an authorised red team exercise on the NSA’s networks. It was not a real attack. However, the speed and scale of the breach are said to have exceeded anything previously observed using conventional methods.
This revelation puts the 12 June decision to cut off access to Fable 5 and Mythos 5 into context. It may not have been just a simple jailbreak. It was a demonstration that a commercial AI model could autonomously compromise the world’s most heavily protected systems.
The official trigger remains the jailbreak reported by Amazon to the Department of Commerce. Facebook’s former head of security, Alex Stamos, says he examined the data and shared Anthropic’s view: ‘valid results but no unique capability justifying a response of this scale. ”
But if Rudd’s account is accurate, it changes everything. An AI model built by a San Francisco start-up, capable of breaching the NSA’s defences in a matter of hours. Not using a known exploit. Through autonomous reasoning.
We’ve gone from ‘AI could become dangerous one day’ to ‘AI has just hacked the NSA during a test’. The question is no longer a theoretical one.
👉 Follow DisruptionMedia to stay up to date with the latest AI news
🚀 THE free newsletter to understand AI, technology and the transformations reshaping the world in less than 5 minutes a day: https://lnkd.in/eN6-tGKB
Credit : DisruptionMedia
