Theoretical computer scientist Scott Aaronson reported on his blog that sources told him major AI labs have begun discreetly testing whether their most advanced models can break important cryptographic protocols. He said the companies would want to get ahead of the issue if their models have that capability.
The claim comes amid growing concern about AI’s potential threat to encryption. Some experts believe that risk could materialize sooner than the threat from quantum computers.
Recent demonstrations have shown AI systems, including Anthropic’s Claude, identifying vulnerabilities in cryptographic systems under review, underscoring AI’s growing capabilities in cryptanalysis.