DeepMind says AI agents found two weaknesses in legacy 1024-bit RSA
Google DeepMind says AI models working under human direction found two weaknesses in legacy 1024-bit RSA, covering signature forgery and cheaper batch factoring of related moduli. Keys of 2048 bits or more remain secure against this specific threat, according to the reporting.
Google DeepMind announced on 6 November that AI models, working under human direction, identified two weaknesses in legacy 1024-bit RSA encryption, according to Crypto Briefing. The researchers named are Dj Dvijotham and Abhradeep Thakurta of DeepMind, working with cryptography experts Nadia Heninger and Emmanuel Thomé.
The first weakness concerns digital signatures, which can be forged under certain conditions. The second is a new algorithm that lowers the cost of batch factoring for related moduli, allowing many to be attacked together more cheaply. The article gives no figures for cost or time savings.
Crucially, modern keys of 2048 bits or longer remain secure against this specific threat, and the report does not suggest systems using stronger keys face immediate risk. Crypto Briefing links no DeepMind paper, so details should be checked against the original publication when available.
DeepMind's broader argument is that AI agents can sharply shorten the time needed to probe the mathematical assumptions behind cryptography, work that has historically taken years of human effort.
Why it matters
Cryptography depends on assumptions that are tested over decades. If AI agents accelerate that testing, weaknesses in legacy schemes may surface faster, which raises the value of knowing exactly which algorithms and key sizes an organisation still uses.
It also shows AI being applied to defensive and offensive security research alike.
Indicops view
Indian banks, government departments and enterprises should inventory any 1024-bit RSA keys, certificates and embedded devices, and retire them in favour of 2048-bit or stronger keys, with post-quantum migration planned alongside. This is a low-regret hygiene task regardless of how the research develops.
More in Research
Back to the newsroomAnthropic study finds AI agent teams do worse when each user brings their own agent
Anthropic's alignment team tested frontier models in four multi-user settings and found group outcomes were worse when each person delegated to a personal agent than when one coordinator served everyone. Three of the environments are released as the MAMUBench benchmark.
DeepMind's Genie 4 turns a single image or prompt into a steerable, audio-enabled world
Google DeepMind has announced Genie 4, a world model that generates explorable environments with native audio and lets users rewrite the scene in natural language while it runs. It is in research preview for trusted testers, with early access planned via Project Genie and the Gemini API.