DeepMind's AI Found Cracks in RSA Encryption — Should You Worry?

By Shakeel Ahmed — AI Growth Lab · November 8, 2026

Reports say Google DeepMind's AI agents found two weaknesses in legacy 1024-bit RSA encryption in a matter of weeks. According to those reports, modern systems using 2048-bit keys or larger are not affected, so your bank account and passwords are safe today. But the work is a warning: the encryption that protects the internet has an expiry date, and AI just made the clock tick faster.

Before the panic starts, here is the honest version. The claim comes from secondary reports, not from a DeepMind publication we can read directly. The results apply only to an old, already-retired key size. And the biggest numbers in the story are theoretical estimates, not measured attacks. This article separates what is reported from what is verified, and explains what it means for you.

What is reported

On November 6, 2026, Google DeepMind announced — according to multiple outlets — that AI agents working under human supervision had helped identify two weaknesses in legacy 1024-bit RSA cryptography. The work was reported to appear in two working papers, published November 5 and 6, titled "Modular Roots from a Restricted Oracle" and "Fan Factories."

The team was reported to be led by DeepMind researchers Dj Dvijotham and Abhradeep Thakurta, working with cryptographers Nadia Heninger of UC San Diego and Emmanuel Thomé of Inria. The announcement was reported to come from Pushmeet Kohli, DeepMind's VP of Science.

Two important cautions before we go further. First, the papers have not been peer reviewed yet. Second, at least one careful checker — ExplainX — searched for DeepMind's own publication and could not find it, and says the claim should be treated as reported, not confirmed, until primary text is public. That does not mean the reports are wrong. It means we are reading the news through other people's summaries, and every number below belongs to the reporting outlets, not to DeepMind directly.

Sources: ExplainX — DeepMind 1024-Bit RSA Claim: What Is Actually Verified? · AI Parabellum — DeepMind AI Agents Expose Two Weaknesses in Legacy 1024-Bit RSA

RSA in plain English

RSA is the lock on much of the internet. When you log in to your bank, update your phone, or send a message, RSA (or its cousins) helps prove that the other side is who it claims to be and that nobody has tampered with the data.

The "bit" number is the size of the key — roughly, how long the lock's combination is. A 1024-bit key is an older, shorter combination. A 2048-bit key is the modern standard: vastly harder to break, and what nearly every serious system uses today.

Here is the key fact: 1024-bit RSA was already retired years ago. The US National Institute of Standards and Technology (NIST) stopped accepting 1024-bit RSA for generating new signatures back in 2013, under standard SP 800-131A. Security experts have called it unsafe for real-world use for over a decade. That is exactly why researchers use it as a test target — it is complex enough to be meaningful, but safe to attack without touching anything live.

So when reports say DeepMind's AI cracked 1024-bit RSA, they are saying the AI picked a lock that the security world had already thrown away.

Weakness one: forging signatures, much cheaper

A digital signature is how your computer checks that a message, a software update, or a document really came from its claimed source. The first reported result is about forging those signatures.

According to the reports, DeepMind's agents found that the protection offered by RSA's padding — extra data mixed in to make forgery harder — depends on how much of the signed message an attacker controls. Once an attacker controls roughly half of the signed block or more, the padding reportedly stops adding meaningful protection. Forging a signature becomes nearly as easy as if there were no padding at all.

The reports give two numbers for this result, and they are very different kinds of numbers:

• Measured: working in an isolated sandbox against an already-solved 1024-bit benchmark key from earlier academic work, the agents' implementation reproduced the forgery using 3.5 times less CPU time than the previous approach — 391 CPU core-years against 1,380 for the original team.
• Theoretical: for harder cases, such as a short 32-byte fingerprint, a new algorithm reportedly improves on the previous best method by about 1,600 times. That figure is a theoretical estimate from the math, not a timed attack.

The distinction matters. A 1,600-times improvement sounds like the lock is broken. In reality, it is an estimate that the hard cases could be much cheaper to attack — not a demonstration that anyone has done it.

Sources: The Library — 1,600 Times Faster, on a Lock Nobody Uses Anymore: What DeepMind's RSA Result Actually Proves · Vanda Team — DeepMind RSA-1024 Weaknesses: What Data Teams Should Fix

Weakness two: cracking many keys at once, cheaper

The second reported result is about batch factoring — breaking many RSA keys at the same time instead of one by one.

Because RSA keys usually share a standard size (1024 or 2048 bits), an attacker going after many keys can, in theory, spread one large upfront computation across all of them, lowering the average cost per key. According to the reports, DeepMind's agents discovered 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.

Nobody has built the "fan factory" this algorithm describes, and the 30-times figure is again an estimate, not a measured attack. But the direction is what worries cryptographers: AI agents are not just running known attacks faster. They are finding new attacks that humans had not thought of.

The reports also say the agents found and fixed four problems in the algorithm's correctness on their own, without another pass from a human expert. That is the part of the story that goes beyond RSA. AI agents are starting to do original cryptanalysis — the science of breaking codes — and doing it in weeks.

Is my data safe?

For almost everyone reading this, the short answer is yes — today.

• Your bank, your email, your messages: these use 2048-bit keys or larger, and often newer systems entirely. DeepMind itself says — according to the reports — that these specific findings do not affect them.
• Your passwords: passwords are protected by different systems (hashing, not RSA), and this research does not touch them.
• Software updates: the signing systems behind your phone and computer updates use modern key sizes. They are not affected by these specific findings.

Who should actually pay attention? Anyone still running 1024-bit RSA anywhere — old servers, legacy industrial systems, forgotten certificates. If your company or your client still has 1024-bit keys in production, the message is simple: retire them now. They were already unsafe before DeepMind's announcement. This is just one more reason.

The longer answer is less comfortable. Every encryption standard eventually falls. RSA-1024 was the standard of the early 2000s; RSA-2048 is the standard of today. The question is not whether 2048-bit keys will one day be weak — it is when. AI that can find new attacks in weeks shortens that timeline, and the security world knows it.

Sources: OfficeChai — Google DeepMind Says It Has Used AI Agents to Break Legacy RSA-1024 Encryption Faster Than Anyone Thought Possible · ExplainX — DeepMind 1024-Bit RSA Claim: What Is Actually Verified?

What comes next: post-quantum encryption

The real story behind this announcement is not RSA-1024. It is the race to replace today's encryption before something — a quantum computer, or an AI agent with a new idea — breaks it for real.

Governments and big tech companies are already moving to post-quantum cryptography: new encryption methods designed to survive quantum computers. NIST has been standardizing them for years, and migration is underway at banks, cloud providers, and governments. The DeepMind reports, if confirmed, will speed that work up. Nothing motivates a migration like a demonstration that the old locks are weaker than we thought.

For developers and small businesses, the practical steps are boring but real:

• Check your key sizes. Anything still on 1024-bit RSA should be upgraded to 2048-bit or higher, or moved to elliptic-curve cryptography.
• Watch your vendors. Cloud providers and certificate authorities are handling most of this migration for you — but only if you keep your systems updated.
• Do not panic-buy anything. No vendor is going to sell you a "DeepMind-proof" product this week, and anyone who tries is selling fear, not security.

AI breaking old encryption is not new — researchers have been doing it for decades. What is new is the speed, and the fact that AI agents are now doing the creative part of the work. That same pattern showed up recently in a very different leak: an AI model that handed users other customers' data. Read it here: A Cyber Attack Shut Every Boston Scientific Factory for 8 Days — And Patients Felt It

And for the bigger picture — how AI tools your team already uses can become a security risk — see: Your Team's Secret AI Tools Just Became the Top Way Companies Get Hacked

The bottom line

Reports say DeepMind's AI agents found two real weaknesses in 1024-bit RSA, and the details — if confirmed — are genuinely impressive cryptanalysis. But the lock they picked was already retired, the biggest numbers are estimates, and your 2048-bit keys are not affected. Do not change your passwords. Do not move your money. Do retire any 1024-bit keys still hiding in your systems.

The warning is real, just slower than the headlines. Encryption ages, AI gets faster at finding the cracks, and the migration to post-quantum cryptography is now a matter of when, not if. The smart move is the boring one: keep your systems updated, use modern key sizes, and let the cryptographers do their work in the open — which, to DeepMind's credit, is exactly what publishing the papers is for.

Originally published on AI Growth Lab — https://aigrowthlabshakeel.blogspot.com/