Global ALARMS Over Dangerous AI Hacking Model Mythos

TL;DR

Anthropic's Mythos AI model, too dangerous to release publicly due to its ability to exploit critical infrastructure vulnerabilities, has already been accessed by unauthorized hackers.

Key Points

• 1.Mythos is Anthropic's AI model capable of finding unknown software vulnerabilities in banks, power grids, and governments. Anthropic deemed it too dangerous to release publicly, triggering global alarm — the Bank of England warned it could 'crack the whole cyber risk world open' and Canada's finance minister compared the threat to closing the Strait of Hormuz.
• 2.Mythos has already been accessed by unauthorized users despite restricted release. A hacker collective organized on Discord gained access to the model, raising serious questions about how securely it was protected and who else may have obtained it.
• 3.There is no regulatory body overseeing AI safety approvals, unlike pharmaceuticals. The hosts argue that a drug affecting thousands requires rigorous safety data and regulator sign-off, yet a potentially civilization-altering AI model faces no equivalent federal scrutiny or transparent expert review.
• 4.Anthropic CEO Dario Amodei met with White House Chief of Staff Susie Wilds over Mythos. The meeting occurred while Anthropic battles the Trump administration in court over blacklisting its Claude model, as the US government tries to balance a hardline stance against Anthropic with national security implications of its breakthrough technology.
• 5.The hosts warn AI-enabled hacking could catastrophically undermine banking and consumer security. They cite risks like AI-spoofed voice scams, hacked two-factor authentication, and the ability to drain accounts instantly — comparing it to the still-unsolved Nigerian scam problem that has persisted for three decades despite regulatory efforts.