DeepMind AI Agents Find Two Weaknesses in Legacy 1024-bit RSA
· Cortex Solutions
On Nov 6, 2026, Google DeepMind said human-guided AI agents found two weaknesses in legacy 1024-bit RSA. 2048-bit keys stay safe. What it means for businesses.
Google DeepMind announced on Friday, November 6, 2026, that AI agents working under the direction of human researchers uncovered two weaknesses in legacy RSA cryptosystems that use 1024-bit keys. The announcement came in a thread on the company's X account, quoted verbatim by Digg: "Guided by humans, our models uncovered two major weaknesses in legacy 1024-bit RSA cryptosystems." CryptoBriefing and RuntimeWire covered the news the same day, followed by DataStudios with a wider analysis of the technical details.
Who is behind the research?
According to CryptoBriefing, the work was led by DeepMind researchers Dj Dvijotham and Abhradeep Thakurta, in collaboration with cryptography experts Nadia Heninger and Emmanuel Thomé. Pushmeet Kohli, DeepMind's VP of Science, said in a post quoted by Digg that the team did the work "in just weeks" alongside "leading cryptanalysis experts."
Weakness one: forging digital signatures
The first result concerns the padding used in digital signatures with 1024-bit keys. According to DataStudios, the weakness appears when the attacker controls part of the signed data, such as a 32-byte fingerprint. In that setting, DeepMind estimates the new method is about 1,600 times better than the previous best known method, which is a theoretical estimate. What was actually run was a repeat of a forgery attack on a benchmark key already broken in an earlier academic study, inside an isolated environment, using about 3.5 times less computing time than the original team. DataStudios notes that DeepMind did not name the affected padding schemes.
Weakness two: factoring many keys at once
The second result is a new algorithm that lowers the cost of factoring large numbers of 1024-bit keys in a batch. According to DataStudios, the cost of each additional key becomes about 30 times lower than the best previous batch method, in a working paper titled "Fan Factories" dated November 5. The site stresses that the figure is a leading-order estimate and that the algorithm has not yet been implemented. CryptoBriefing described the result as making an attack on a group of related keys cheaper.
How did the AI work here?
According to DataStudios, the team used a two-agent setup: one agent generates mathematical proofs and a second verifies them, with some identities machine-checked in the Lean 4 proof language. CryptoBriefing says the agents did much of the exploratory heavy lifting, while humans still steered the research.
Is your current encryption at risk?
The short answer from DeepMind: no, if you use modern keys. Kohli said these results "do not affect the security of modern encryption schemes such as RSA 2048." CryptoBriefing and RuntimeWire both state that 2048-bit and larger keys remain safe against these specific attacks. Digg also noted that DeepMind's posts did not name specific affected systems and did not describe attacks in use.
The deeper message came from Kohli himself, who said the results show why "we must not assume" the mathematical "hardness" of cryptographic schemes, and how AI "can help defenders improve security standards." According to DataStudios' analysis, 1024-bit keys still exist in older hardware, embedded systems, older smart cards and long-lived certificates.
What it means for Saudi businesses
This section is the editorial view of the Cortex team, not part of DeepMind's announcement:
- Inventory the encryption keys and certificates in your systems and look for any 1024-bit RSA key, especially in older devices and internal systems that have not been updated in years.
- Ask your system and device vendors which key lengths they use for signing and encryption, and request a clear plan to move to 2048 bits or more.
- Do not treat encryption as a one-time decision. AI is speeding up the discovery of weaknesses, which means security settings need regular review.
- Start planning early for quantum-resistant encryption, because migration in large systems takes time.
The good news is that this research was announced publicly and for defensive purposes, giving businesses a chance to act before theoretical weaknesses turn into practical attacks. If your company runs older custom-built systems that need updating or rebuilding to modern standards, the Cortex software development team can help: https://cortex.sa/services/custom-software
Sources: Google DeepMind and Pushmeet Kohli posts on X as quoted by Digg (November 6, 2026), CryptoBriefing (November 6, 2026), RuntimeWire (November 6, 2026), and DataStudios (November 2026).