DeepMind’s AI Agents Find Faster Ways to Break Legacy RSA-1024
Key points
- Google DeepMind says AI agents found cryptanalysis techniques that weaken legacy 1024-bit RSA by up to roughly 1,600 times versus the prior best attack, announced by chief scientist Pushmeet Kohli 1.
- The first finding shows that once an attacker controls about half or more of a signed data block, padding gives virtually no extra protection against signature forgery 1.
- The second finding cuts the average cost of factoring each additional 1024-bit key in a batch by about 30 times compared with earlier batch methods 1.
- DeepMind stresses that today's 2048-bit and larger keys, and the systems behind banking, messaging and software updates, remain unaffected 1.
A Speedup Aimed at Old Keys
Google DeepMind researchers report that AI agents have uncovered cryptographic techniques that expose weaknesses in legacy RSA encryption up to 1,600 times faster than earlier approaches. The announcement came from Pushmeet Kohli, DeepMind’s Vice President and Chief Scientist, in a social media post 1.
The work targets 1024-bit RSA, a standard that has largely been replaced by keys of 2048 bits or more. The study was led by Dj Dvijotham and Abhradeep Thakurta, with collaborators from the University of California San Diego and France’s Inria 1.
Padding Fails Under Attacker Control
The first result concerns digital signatures, which let computers confirm that software updates, websites and documents come from legitimate sources. These schemes usually add preset data called padding before signing, which is understood to block forgery. What had not been fully quantified was how well it holds up when an attacker can influence part of the signed message 1.
The agents found that when an attacker controls roughly half of the signed block or more, padding adds almost no protection, and forging a signature becomes no harder than attacking an unpadded system. To check this, the team ran the method in a sandbox against an already-solved RSA-1024 benchmark from earlier academic work. The AI-generated implementation reproduced the result with 3.5 times less CPU time, and with less than one-thousandth of the compute needed to factor the benchmark key itself 1.
Even when the attacker controls only a small slice, such as a 32-byte digital fingerprint, the agents devised an algorithm that in theory improves on the previous best attack by about 1,600 times for 1024-bit keys 1.
Cheaper Batch Factoring
The second finding involves factoring many RSA keys of the same bit length at once. Because many deployments use standardized key sizes, an attacker could in theory spread the cost of heavy computation across numerous keys 1.
The AI-discovered algorithm improves the asymptotic efficiency of this process. It lowers the average cost of factoring each additional 1024-bit key by roughly 30 times compared with earlier batch techniques 1.
What It Means for Security
DeepMind is explicit that the findings do not threaten the encryption protecting online banking, messaging apps, software updates or digital currencies. Instead, they show how AI can sharply speed up the testing of the mathematical assumptions beneath digital security 1.
Cryptanalysis, the practice of hunting for weaknesses before bad actors do, has historically moved over years or decades. The company says its agents compressed that timeline by quickly exploring problems that usually demand extensive manual research. Kohli argued the results underline the need to keep reassessing cryptographic assumptions as AI grows better at advanced mathematics, and the researchers see AI-powered cryptanalysis as a tool for stress-testing future systems before new threats emerge 1.
Companies mentioned: DeepMind • Google (GOOGL $284.78 ▲0.8%)
Primary sources
1,600x faster: Google DeepMind agents expose weaknesses in older encryption methods (interestingengineering.com) – This Interesting Engineering report covers DeepMind's announcement, made by Pushmeet Kohli, that AI agents found new attacks on legacy 1024-bit RSA. It describes two findings: padding offers almost no protection against forgery once an attacker controls about half of a signed block, with a theoretical roughly 1,600x improvement even for a 32-byte fingerprint, and a roughly 30x cut in per-key cost for batch factoring. It also covers the sandbox validation (3.5x less CPU time on a solved benchmark), the research team (Dvijotham, Thakurta, UC San Diego, Inria), and the assurance that modern encryption is unaffected. Published by interestingengineering.com.
Powered by News Ranker