Science & Tech · AI · 9 hrs ago
Google DeepMind Agents Find Cheaper Attacks on Legacy RSA-1024
Google DeepMind’s AI agents found cheaper ways to attack 1024-bit RSA systems, working with researchers from UC San Diego and Inria.
RSA is a widely used encryption method, and 1024-bit keys are now considered obsolete.
One finding concerns digital signatures: under certain conditions, an attacker who controls about half of the signed data can overcome padding that is meant to make forgery harder.
In a sandboxed test, the agents cut the computing time for an earlier signature-forgery attack from 1,380 to 391 CPU core-years.
The attack does not recover a private key and requires about 4 billion signing queries in advance; it also does not work on devices that hash messages internally.
A second, theoretical method could make factoring each additional 1024-bit RSA key about 30 times cheaper, but it has not been implemented and assumes a costly initial calculation.
Modern systems using 2048-bit or larger keys remain secure, according to Google DeepMind, while the researchers say AI can help test assumptions about cryptographic security.
Google DeepMind says AI agents working with human supervision found ways to reduce the cost of attacking legacy 1024-bit RSA systems.
In a test, the agents cut the computing time for an earlier signature-forgery attack by 3.5 times.
The research also describes a theoretical method that could make factoring each additional 1024-bit RSA key about 30 times cheaper on average.
The researchers said the findings do not affect modern encryption protocols, which almost universally use keys of 2048 bits or larger.
- Who
- Google DeepMind researchers, working with cryptographers from UC San Diego and Inria.
- What
- AI agents found methods that reduce the cost of attacks on 1024-bit RSA systems.
- When
- The research was published on 2026-11-09.
- Where
- Not stated.
- Why
- To test assumptions about the security of cryptographic systems and assess how AI can support mathematical research.
This story does not have two clearly opposing sides.
An academic forgery attack on a 1024-bit RSA key required 1,380 CPU core-years.
Two AI agents designed, ran and debugged a forgery computation, with about 20 messages from a human researcher.
SecurityWeek published the report on the research findings.
- Forgery test
- 391 CPU core-years, compared with 1,380 in the earlier study
- Agents’ work
- Two agents; under four days; approximately 20 researcher messages
- Key factoring estimate
- About 30 times cheaper per additional 1024-bit key on average
- Estimated cost to factor one key
- 500,000 to 1 million core-years
- Modern key sizes
- 2048 bits or larger, according to Google DeepMind
What people are saying
In practice, these results only affect obsolete 1024-bit RSA systems, and modern encryption protocols, which almost universally use 2048-bit or larger keys, remain secure.
Once they control roughly half of the block or more, padding adds no protection at all.
Together, these results show that in a new era of AI-enabled mathematics and computation, we cannot treat cryptographic hardness as a fixed assumption. Using AI to rigorously test our previously held assumptions will ensure that modern encryption stays secure,