What was found
Google DeepMind used AI agents in cryptanalysis to produce two new results against legacy RSA-1024 encryption. Chief Scientist Pushmeet Kohli announced the findings on social media. The work shows that agents operating under human supervision can accelerate mathematical research that traditionally takes years.
Two separate results
The first finding concerns digital signatures. Researchers found that "padding" — extra preset data added before signing — offers virtually no protection once an attacker controls roughly half of the signed block. The second targets batch factoring of multiple RSA keys of the same bit length: the new algorithm cuts the average computational cost of each additional key by about 30 times versus previous batch techniques.
Modern encryption unaffected
The research focuses on 1024-bit RSA, a standard largely replaced by 2048-bit and larger keys. According to DeepMind, the findings do not threaten encryption used for online banking, messaging apps, software updates or digital currencies.
Why it matters
Cryptanalysis is the science of testing encryption systems for weaknesses before malicious actors exploit them. The researchers argue that as AI becomes more capable at advanced mathematics, cryptographic assumptions must be continuously reassessed. The study was led by Dj Dvijotham and Abhradeep Thakurta in collaboration with researchers from UC San Diego and France's Inria.
