Trump Halts Export of AI Models That Can Crack Codes

The Trump administration has taken a decisive step to halt the export of advanced AI models that possess the capability to crack security codes. The decision targets new generation AI systems, specifically mentioning Anthropic's Mythos and Fable, which are described as "super-hacking AI." These models are deemed dangerous because they can autonomously identify security flaws in commonly used software.

This move raises significant concerns about the current state of computer security. Questions are emerging about whether bank accounts, mobile phones, and computers are still safe from sophisticated AI-driven attacks. The development prompts contemplation of a future where machines could potentially break all digital locks, and governments would control who gains access to this powerful technology.

To address these pressing issues, The ZEIT spoke with Jon Callas, a renowned figure in the cybersecurity world. Callas is credited with co-inventing the PGP email encryption standard and developing Apple's hard drive encryption. He also advises numerous companies and governments globally on computer security matters. His insights are crucial in understanding the implications of these powerful AI tools and the potential risks they pose to digital infrastructure.

“

Mythos and Fable of Anthropic are extremely dangerous because, as so-called super-hacking AI, they can independently discover security vulnerabilities in widely used software.