Matthew Green Warns AI Could Break Public-Key Encryption

Cryptographer Matthew Green has raised public alarm over the potential for rapid artificial intelligence progress to destabilize global digital security, warning that current encryption standards may become obsolete far faster than human organizations can replace them. The commentary, highlighted on October 9, 2026, by technology writer Simon Willison, urges the cybersecurity community to look past social respectable norms and confront worst-case possibilities regarding foundational defense mechanisms.

Green outlined two specific probabilistic assessments concerning the future of cryptographic resilience. First, he estimated a 1% chance that society functionally resides in Minicrypt—the hypothetical world named by computer scientist Russell Impagliazzo in which public-key encryption is fundamentally impossible. Second, and more critically for current security planning, Green assigned a 15% chance that the software industry will functionally lose confidence in its existing public-key encryption algorithms under pressure from rapid AI breakthroughs.

The core vulnerability facing infrastructure maintainers lies in an order-of-magnitude speed disparity. Artificial intelligence models are producing surprising capabilities at a rapid pace, whereas human institutions update and formalize cryptographic standards at a notoriously sluggish rate. Green emphasized that even if human engineers rely on the very best AI assistance, the process of replacing entrenched global standards cannot keep up with real-time automated discoveries. Recovery from an unexpected algorithmic collapse is only possible if organizations do the necessary preparation work in advance.

For security practitioners and software architects, these observations highlight an urgent need to build flexible, adaptable cryptographic pipelines. Relying on traditional standards without a fallback strategy leaves critical infrastructure exposed to sudden model developments. As artificial intelligence advances into automated research and exploitation, waiting for an active compromise before upgrading public-key algorithms will leave organizations unable to bridge the gap between AI surprises and human response times.