Google DeepMind Says AI Agents Found Two Flaws in Old RSA Encryption
By Adam, AI Engineer · 9 November 2026 · 3 min read
Google DeepMind says its AI agents, guided by human experts, found two weaknesses in an old form of RSA encryption that uses 1024-bit keys. The lab announced the research on X on 6 November, and it was reported by Crypto Briefing and Runtime Wire.
DeepMind is clear that this does not break the encryption most of the internet uses today.
What DeepMind found in RSA 1024
RSA is a common way to lock data and to sign digital documents. The longer the key, the harder it is to break. According to the coverage, the two findings are:
- Fake signatures: a weakness that can let someone forge digital signatures made with old 1024-bit keys.
- Cheaper bulk attacks: a new method that lowers the cost of attacking many related RSA keys at the same time.
Crypto Briefing reports that the work was led by DeepMind researchers working with outside cryptography experts. The AI agents did much of the trial-and-error work, while humans set the direction.
Modern keys are still safe, DeepMind says
DeepMind says modern RSA with 2048-bit or longer keys remains safe from these specific attacks. Most websites and services moved away from 1024-bit keys years ago. The lab says the bigger point is that encryption standards need constant testing, and AI can make that testing much faster.
Why it matters
This is a warning about old systems, not a crisis. In our analysis, the risk sits in forgotten places: old routers, internal tools, legacy software or signing keys created many years ago that still use 1024-bit RSA. If AI makes attacks on those keys cheaper, they become easier targets. A simple step for any business: ask your IT team or provider whether anything you run still uses 1024-bit keys, and plan to replace it with 2048-bit or stronger.