Hopp til hovedinnhold

DeepMind: AI agents find two weaknesses in legacy RSA-1024 — 2048-bit stays safe

Google DeepMind says human-guided AI agents exposed signature forgery and cheaper batch factoring against legacy 1024-bit RSA. Modern 2048-bit keys remain unaffected, according to the lab.

Av Joachim HøgbyKilde: Google DeepMind

Google DeepMind says AI agents under human guidance uncovered two material weaknesses in legacy 1024-bit RSA — with lower compute budgets than previously assumed. Modern protocols that use 2048-bit or larger keys remain secure against these findings, according to DeepMind.

The announcement came via DeepMind VP of Science Pushmeet Kohli and the official DeepMind account on Friday 6 November 2026. Researchers stress that RSA-1024 is used as a safe proving ground, not as a judgment on current production cryptography.

Signature forgery and cheaper batch factoring

The first result concerns digital signatures — how systems verify that software updates, documents or websites truly come from the claimed source. Many standards add padding before signing. DeepMind says protection depends on how much of the signed block an attacker controls: once they control roughly half or more, padding adds no protection against mathematical shortcuts.

In a sandboxed simulation — without touching live systems — the agents reproduced a result on an already-solved 1024-bit benchmark key from earlier academic work with 3.5× less CPU time, and at less than a thousandth of the compute needed to break the key itself. When an attacker controls less of the block, such as a short 32-byte fingerprint, the team found a new algorithm that, by DeepMind’s theoretical estimates, improves on the previous best method by roughly 1,600× for 1024-bit keys.

The second result concerns factoring many numbers of the same size at once. Because RSA keys typically share a standard bit length, an attacker can in theory spread one large upfront computation across many keys. DeepMind says its agents discovered a new algorithm with better asymptotic complexity than previously known batch methods, cutting the average cost of factoring each additional 1024-bit key by about 30× versus prior batch approaches.

Led by DeepMind Science with external cryptographers

According to Kohli, the work was led by Dj Dvijotham and Abhradeep Thakurta, in collaboration with Nadia Heninger (UC San Diego and Inria) and Emmanuel Thomé (Inria). DeepMind points to two technical papers on digital signatures and key factorisation hosted on Google Cloud Storage. The broader message: cryptographic “hardness” is not a fixed assumption in an era where AI agents can probe mathematical assumptions faster than before.

Why it matters for leaders

  • Inventory legacy RSA-1024 — do not panic about 2048+. DeepMind limits the findings to obsolete 1024-bit systems. If you still run 1024-bit certificates, VPN or HSM setups, or older device keys, schedule migration — not because 2048 “just fell”, but because hardness margins can shrink faster than many roadmaps assume.
  • Signature chains and update channels. Signature forgery is about trust in what is “genuine”. Security and IT should map where padding and signing-oracle access (signing services, HSM APIs) still exist, and whether older padding variants remain in use.
  • Vendor requirements. Ask vendors to document minimum key sizes and migration plans. For regulated sectors this is also a compliance question against established stronger-key requirements.
  • AI changes the tempo of cryptanalysis. DeepMind’s point is that defenders should stress-test hardness assumptions more proactively. That is a strategic architecture risk, not a same-day patch event for modern TLS.

Sources and media

Illustration: AI-generated with Grok Bot / hogby.ai

Mer om AI

Oppskriften. Strategien. Hver dag.

AI-nyheter for ledere. Kuratert av en CIO som bygger det selv. Daglig i innboksen.