Anthropic CEO Dario Amodei reportedly revealed firms testing Mythos warned it was a brilliant weapon and mentioned utilizing it ought to require a “gun license,” based on submit on X by Polymarket (which claims to be the world’s largest decentralised prediction market that enables customers to wager on the outcomes of real-world occasions utilizing cryptocurrency). NDTV couldn’t independently confirm this assertion or the place Amodei mentioned this.
Fable 5 is constructed on Anthropic’s underlying mannequin Mythos, which the corporate previewed in April and stored it away from mass launch as a result of it was deemed too highly effective amid issues that unhealthy actors may use it to hack crucial infrastructure resembling banking methods or construct bioweapons. In response to the corporate, Mythos was in a position to establish flaws in each main working system and internet browser it examined, a few of these vulnerabilities had been reportedly mendacity undetected for many years. Therefore, the corporate launched a managed program referred to as Undertaking Glasswing, sharing it with round 50 vetted organisations, together with Google, Apple, Amazon, Microsoft, and CrowdStrike, to utilise for defensive cybersecurity work.
Additionally Learn | Opinion | The AI Kill Switch, And Why Sovereign AI Just Became Non-Negotiable
Final Tuesday, the corporate launched a model of Mythos with guardrails to dam responses in high-risk areas like cybersecurity and biology. If somebody tried to utilise the newest model Fable 5 to hack right into a crucial system or construct a bioweapon, the mannequin would routinely slip again into the sooner less-capable model referred to as Claude Opus 4.8.
In response to Vals AI, which tracks AI mannequin efficiency, Fable 5 ranked as probably the most succesful publicly obtainable AI mannequin on its benchmark checks.
Anthropic has been in lively talks to determine the scenario which it calls a “misunderstanding.”
The US authorities on Friday ordered Anthropic to immediately revoke access for overseas nationals (together with these employed by Anthropic) to 2 of its strongest AI fashions – Claude Fable 5 and Claude Mythos 5 – citing nationwide safety issues. To conform, Anthropic went forward and suspended entry for all. The directive was to revoke entry for all overseas nationals, whether or not contained in the US or exterior.
Additionally Learn | Why Anthropic Disabled Its Most Powerful AI Model Days After Launch
Anthropic put out a prolonged statement during which it mentioned: “The online impact of this order is that we should abruptly disable Fable 5 and Mythos 5 for all our prospects to make sure compliance.” It added it believes “it is a misunderstanding” and the corporate is working to revive entry “as quickly as attainable.”
The federal government’s “export management directive” to Anthropic got here amid issues of a possible “jailbreak.” Within the assertion it launched, Anthropic mentioned, the federal government has supplied solely verbal proof of a “potential slender, non-universal jailbreak”. Merely put, a jailbreak is a method used to bypass an AI mannequin’s built-in security restrictions and acquire responses it will usually refuse to supply.