Google DeepMind has turned its AI agents loose on one of the oldest locks in internet security — and the agents came back with two new ways to pick it. The lock is 1024-bit RSA, an encryption standard the industry retired years ago. The headline number is startling: in one case the AI-found method works around 1,600 times faster than the best previous approach. But before anyone panics, DeepMind is clear about the limits — modern encryption, the 2048-bit keys that protect online banking, messaging apps and software updates, is not affected.
The research was announced on Friday 6 November 2026 by Pushmeet Kohli of Google DeepMind, and written up in two technical papers. It is a demonstration of what AI can now do in cryptanalysis, the science of breaking codes — and a warning that old systems still running retired encryption should be upgraded while there is still time.
What DeepMind’s AI agents actually did
Two AI agents, guided by about 20 short messages from a human researcher, reran a recent attack on 1024-bit RSA signatures in under four days — using about 3.5 times less CPU time than the academic team that first pulled it off. One model acted as supervisor, writing the design, setting checks for each stage and auditing the results. A second model wrote and ran the code, fixing failures as they came up. Between them they produced around 33,500 lines of new code and launched 336 distributed experiments, peaking at roughly 300,000 CPU cores and 16 graphics processors. Along the way the agents even tracked down a 32-bit index overflow that had made their first computed answer wrong.
The benchmark they attacked was a 1024-bit instance first broken by a team led by Laura Shea, Miro Haller, Adam Suhl, Nadia Heninger and Emmanuel Thomé, whose paper went up on the IACR ePrint archive in September 2026. That team spent 1,380 CPU core-years over five calendar months, using a hardware security module as the signing oracle. DeepMind’s agents, using the same algorithm, polynomials and bounds, computed the first root in 391 core-years. The paper credits the saving mainly to better code rather than better hardware, and says the much shorter calendar time “reflects mainly parallelism”.
The work was led by Dj Dvijotham and Abhradeep Thakurta, working with outside cryptographers Nadia Heninger of the University of California San Diego and Inria, and Emmanuel Thomé of Inria. DeepMind says it has published full technical papers on both results so the wider security community can check them and build on them.
The first weakness: faking digital signatures
The first result asks how much protection the padding in a digital signature really gives. When a device signs a document with RSA, it wraps the data in padding — extra bytes designed to make forgery hard. DeepMind’s agents found that the answer depends on how much of the signed block the attacker controls.
Through the PKCS#11 mechanism CKM_RSA_PKCS, a caller who can supply 65 of the 128 bytes in a 1024-bit block faces a problem no harder than raw, unpadded signing — the padding, in effect, stops adding protection. A 64-byte digest such as SHA-512 behaved about the same in their measurements.
The harder case is a short 32-byte fingerprint like SHA-256, where the attacker controls only a small slice of the block. For that case the agents produced a new algorithm that, on DeepMind’s theoretical estimates, improves on the best previously known method by roughly 1,600 times for 1024-bit keys. That figure needs careful reading: it is a theoretical estimate, not a measured result, and the paper says it is asymptotic and was not run at this size. The measured result is the 3.5 times CPU saving on the solved benchmark key.
There are also hard limits on the attack. An attacker needs direct calls into the signing interface — about four billion of them in the reproduced computation, all made before the target was known. And mechanisms that hash the message inside the device, such as CKM_SHA256_RSA_PKCS, never let the caller choose the block at all, so they fall outside both attack models. Current guidance already no longer approves 1024-bit RSA for generating signatures.
The September paper that started all this had already pushed the point further than 1024-bit keys: with a signing oracle, its authors concluded, RSA’s concrete security should be 15 to 30 bits lower than factoring-based estimates for keys from 1024 to 4096 bits. They wrote: “Even 4096-bit RSA does not appear to meet a 128-bit security level in this attack model.”
The second weakness: cracking many keys at once
The second paper targets a different problem: factoring many RSA keys at once. When the shared precomputation costs about one factorisation, Coppersmith’s multiple number field sieve has set the best known cost for each extra integer since 1993. DeepMind’s new method, nicknamed the “fan factory”, gives each target its own family of number fields fed by that shared work. At leading order, each further 1024-bit integer would cost about one fifty-eighth of a standard number field sieve run — around 30 times less than Coppersmith’s method.
One model generated the proofs and a second checked them, and the algebraic identities were machine-checked in Lean 4. The authors are clear that no fan factory has actually been built — this result is, for now, on paper only.
What this means for you — and for Pakistan
For ordinary users, the message is reassuring: nothing in these findings threatens online banking, messaging apps, software updates or digital currencies. No live keys were broken — the experiments ran in a sandbox, against a standard that the United States standards body NIST stopped accepting for generating new signatures back in 2013. Modern protocols almost universally use 2048-bit keys or larger, and DeepMind says those remain secure.
The real warning is for the old systems still out there. Legacy servers, ageing hardware and outdated software can still be running 1024-bit RSA, and the research shows that AI has now become a working tool for cryptanalysis — capable of finding sharper attacks, faster, with only light human guidance. For banks, government departments and companies in Pakistan still running legacy systems, the message is simple: audit what encryption your old systems use, and retire anything still on 1024-bit keys.
Google’s AI security work is moving fast — from Google’s Gemini 4 Carbon, a model built for cybersecurity, to OpenAI’s GPT-6.2 Astra — and this DeepMind result shows the same technology can now probe the foundations of internet security itself. The locks most of us rely on still hold. The ones we retired years ago, it turns out, were weaker than the textbooks assumed.
Sources: Google DeepMind finds 1,600x faster way to expose encryption flaws; Google DeepMind’s AI Agents Find Two Weak Spots in Legacy 1024-Bit RSA
Related Posts
TechnologyPTA and Meta Block 176,000 WhatsApp Accounts Linked to Recycled Numbers in Pakistan
TechnologyWhatsApp to Let You Hide Your Phone Number: New Privacy Setting Spotted in Beta
TechnologyValore Quad Electric Vehicle Launched in Pakistan With 292 KM Range — Price Still a Mystery
TechnologySix Satellite Internet Companies Register in Pakistan — But Starlink’s Licence Still Waits on Security Clearance
TechnologyVivo V80 Price in Pakistan Leaked Ahead of Launch — Rs209,999 Tipped for 256GB Model
TechnologyOpenAI Raises $11 Billion at $1.25 Trillion Valuation — Biggest AI Funding Round YetMost Popular
How-ToPTA Tax Calculator 2026: Check Mobile Tax on Any Phone418 views
Pakistan NewsPIA Boeing 787 Lease Deal: What It Means for the Airline391 views
Canada JobsTech Jobs in Canada: Roles, Salaries, and Skills Needed388 views
SportsUEFA Champions League 2026-27: Format, Teams and Schedule385 views
TechnologyPM Shehbaz Sets November Deadline for Alibaba AI and Digitalisation Projects in Pakistan366 views
ShowbizWomen’s Fashion in Pakistan 2026: Trends, Brands and Style Guide361 views
