On November 6, Google DeepMind announced that its AI agents — working under human supervision — found two new weaknesses in legacy 1024-bit RSA, one of the oldest encryption standards on the internet. The research, shared by DeepMind VP of Science Pushmeet Kohli, is a striking demo of AI doing real cryptanalysis. But before anyone panics: your bank account is fine. Here is what actually happened.
What the Agents Actually Found
DeepMind's team, led by Dj Dvijotham and Abhradeep Thakurta with outside cryptographers Nadia Heninger of UC San Diego and Inria and Emmanuel Thomé of Inria, published two technical papers. Both target 1024-bit RSA — a standard the industry retired years ago, and one the US standards body NIST stopped accepting for new signatures back in 2013. The experiments ran in a sandbox against a retired standard, so no live keys were broken and nobody's data was at risk.
Crack One: Forging a Signature
The first result is about digital signatures. DeepMind's agents found that when an attacker controls roughly half or more of the signed block, the padding that is supposed to protect the signature stops adding meaningful protection — forging one becomes nearly as easy as if there were no padding at all. The agents reproduced a recent academic forgery using 3.5 times less CPU time than the original team, and at under a thousandth of the computing power it would take to factor the underlying key itself. For the harder cases — where the attacker controls only a small slice, like a 32-byte fingerprint — the agents produced a new algorithm that DeepMind estimates improves on the best previously known method by roughly 1,600 times. That 1,600x figure is a theoretical estimate, not a measured result, and the distinction matters.
Crack Two: Factoring Keys in Bulk
The second paper attacks a different angle: breaking many RSA keys at once. The agents found a new algorithm with better asymptotic complexity than any previously known batch method, cutting the average cost of factoring each additional 1024-bit key by about 30 times. On its own, factoring one 1024-bit key is still an enormous job. But at scale — across thousands of old keys — the economics shift.
Why Your Bank Account Is Safe
DeepMind is explicit about the limits, and they matter more than the headlines. These results apply only to obsolete 1024-bit RSA. Modern systems almost universally use 2048-bit keys or larger, and DeepMind says those are unaffected. Online banking, messaging apps, software updates and digital currencies are not threatened. The 1024-bit standard was chosen as a proving ground precisely because attacking it endangers nobody — complex enough to be a meaningful test, retired enough to be safe.
Why This Still Matters
The real story is not the old lock — it is the locksmith. AI agents just did publishable cryptanalysis research in days, guided by about 20 short messages from a human researcher. That is a new capability: AI systems that can stress-test the math our security rests on, faster than human teams alone. It comes in a week when AI security is suddenly everywhere — OpenAI shipped
GPT-6 Cyber, a model built to find vulnerabilities, and
Anthropic gave security teams a way to watch AI agents inside isolated VMs. The labs are racing to use AI to find the cracks before someone else does. And
Grok's new invisible watermarking shows the same defensive instinct on the content side.
The lock that matters — modern encryption — still holds. But the message from DeepMind is clear: the tools that test our locks are getting dramatically faster, and the security world needs to keep up. Old systems should be retired, not trusted — and the next generation of encryption needs to be ready for AI-powered attackers, not just human ones.
Comments
Post a Comment