SecurityOct 27, 2026

AI Threats to Encryption May Not Need Quantum Computers

AI is making mathematical attacks more practical, but standards-grade encryption remains safe for now.

Why AI Is Changing the Threat Model

What the Latest Research Really Shows

What Developers and Security Teams Should Do

For years, quantum computers were cast as the existential threat to public-key cryptography. The newer, quieter risk is that artificial intelligence could help researchers find better classical attacks on the math underneath encryption. Public-key systems depend on problems that are computationally hard to solve; if AI accelerates the search for better algorithms, the security assumptions these systems rely on can erode even without a quantum machine.

OpenAI's reported result on the Exponential Time Hypothesis is a dramatic example, but not a practical break. If confirmed, it suggests a general method that avoids exponential growth for certain hard decision problems — a theoretical breakthrough. Yet an asymptotic improvement can still be too slow to break real keys. What matters for defenders is the pattern: AI is becoming a tool for mathematical discovery, which lowers the cost of exploring attacks that were previously impractical.

The Anthropic HAWK research shows this pattern more concretely. With roughly $100,000 in model usage, researchers recovered a signing key in a small experimental version of a proposed post-quantum scheme, leading its developers to withdraw it. That does not mean finalized standards like ML-KEM are broken; NIST says its post-quantum standards are unaffected. It does show how AI-assisted analysis can expose weaknesses before a scheme becomes entrenched. The BSI advice on Classic McEliece is separate and should not be conflated with an AI-driven break.

For developers and security teams, the takeaway is not panic — it is preparation. Keep cryptographic dependencies current, track NIST and BSI guidance, and build in the ability to swap algorithms if a transition becomes necessary. The biggest risk from an AI-era encryption scare may be overreacting to theoretical headlines or ignoring the underlying need for cryptographic agility. Watch the research, but do not abandon working standards before there is a demonstrated attack.

Plan for cryptographic agility now, but don't treat every AI research headline as an active breach.

#ai encryption threat#post-quantum cryptography#encryption security#ai safety
View source ↗