The "doomsday AI" that can't be released to the public is keeping many important people awake at night.
Scott Bessent called a security alert to Wall Street leaders.
Lagarde (ECB) said that "it could be very bad if it falls into the wrong hands."
Bailey (BOE) woke up thinking about what he did wrong in a past life.
Why did an AI still in testing stop the G-7?
- The Meeting That Scared Wall Street
US Treasury Secretary Scott Bessent met with market leaders to warn them: Mythos could usher in an era of autonomously launched cyberattacks by bots.
The Treasury itself is seeking emergency access to the model to map the risks before it falls into the wrong hands.
"I feel confident that everyone is rowing in the same direction to build resilience," Bessent told CNBC.
- Human-Free Attacks in the Loop
Mythos has not yet been released to the public. Anthropic has only allowed testing with a restricted group of companies (Project Glasswing).
The model can autonomously identify and exploit cyber vulnerabilities. Without a human in the process.
John Williams, president of the New York Fed: the advances "have moved much faster than many expected."
- Europe Left Out of the Briefing
The extent to which Anthropic will share details of Mythos with authorities and banks outside the US remains unclear.
An American leader of a European bank said that "a briefing was promised." Nothing more than that.
At the G-7, ministers discussed creating a framework for AI governance. The path forward is still undefined.
4 Any hostile country can do the same.
If an American company built Mythos, a company in an opposing country can too. Without warning the White House.
The conflict in the Middle East increases the perception of state-sponsored attacks.
How to survive in a world of wars where frontier AI capabilities advance faster than the governance frameworks designed to control them?
Anthropic AI keeps giving me the Marvel's Ultron vibes, I hope we don't get there.