Google DeepMind AI finds weaknesses in old RSA encryption: is your data safe?
Google DeepMind's AI agents found two new weaknesses in legacy 1024-bit RSA encryption, including a signature attack up to 1,600x more efficient. Modern 2048-bit encryption is not affected.
Google DeepMind says its AI agents have found two new weaknesses in 1024-bit RSA, an older form of encryption that still exists in some legacy systems. One finding could make forging certain digital signatures about 1,600 times more efficient. The good news: modern encryption, which uses 2048-bit keys or larger, is not affected. Here’s what was found, in simple terms, and whether you need to worry.
What is RSA, and why does key size matter?
RSA is one of the most widely used types of encryption. It helps protect websites, emails, software updates and digital signatures. Its security depends on how hard it is to split a very large number into its two prime factors.
The key size is how big that number is. Bigger keys are much harder to break:
- 1024-bit RSA: old. Security experts have considered it retired and unsafe for years.
- 2048-bit RSA and larger: today’s standard for most websites and services.
What did DeepMind’s AI agents find?
According to OfficeChai and AI Weekly, DeepMind’s agents, working under human supervision, found two weaknesses:
1. Digital signatures are easier to forge than thought
RSA signatures use extra data called padding to make forgery harder. The research found that padding protects less than assumed. If an attacker controls roughly half or more of the signed data, forging a signature becomes almost as easy as if there were no padding at all.
- In a sandboxed test on a 1024-bit key that had already been solved, the agents reproduced the attack using 3.5 times less computing time.
- For short inputs, like a 32-byte fingerprint, they found a new method DeepMind estimates is about 1,600 times faster than the best previous method. That figure is a theoretical estimate.
2. Breaking many keys at once is cheaper
When an attacker wants to break many keys of the same size, a new algorithm lowers the average cost of breaking each extra 1024-bit key by about 30 times. This is called batch factoring.
DeepMind argues that the difficulty of breaking encryption can no longer be treated as a fixed assumption, and that defenders should measure it proactively.
Who did the research?
The team included Dj Dvijotham and Abhradeep Thakurta from DeepMind, plus well-known cryptographers Nadia Heninger (UC San Diego and Inria) and Emmanuel Thomé (Inria). DeepMind says the work is meant to help the security community find weaknesses before attackers do.
Is my encryption safe?
For almost everyone, yes.
- Websites: modern sites use 2048-bit RSA or newer methods like elliptic-curve cryptography. These are not affected.
- Your phone, banking apps and messaging apps: use modern encryption and are not affected.
- Who might be at risk: organizations still running old systems with 1024-bit keys, such as legacy servers, old hardware devices, outdated VPNs, or long-lived certificates and signing keys that were never upgraded.
What organizations should do
- Find any remaining 1024-bit RSA keys in servers, devices, certificates and code-signing systems.
- Replace them with at least 2048-bit RSA, or switch to modern alternatives.
- Check old hardware like routers, cameras and industrial devices, which often keep weak keys for years.
- Plan for post-quantum encryption. Future quantum computers threaten today’s RSA too, so it’s worth planning the move to newer standards now.
Why this matters
This is another sign that AI is changing cybersecurity on both sides. AI agents can now help researchers find mathematical shortcuts that humans missed for years. The same kind of tools can help attackers too, as we’ve seen with LLMjacking and AI-assisted hacking campaigns. Google has also been building AI for defenders, like its Gemini 4 Carbon cybersecurity model.
The lesson for everyone is simple: old security doesn’t stay safe forever. Keeping software and devices updated is still one of the best ways to stay protected.
Frequently asked questions
Did Google DeepMind break RSA encryption?
No. DeepMind’s AI agents found two weaknesses in old 1024-bit RSA that make some attacks much cheaper. Modern 2048-bit RSA and larger keys are not affected.
Is 1024-bit RSA still used?
It has been considered unsafe for years, but it can still be found in some legacy servers, old hardware and outdated certificates that were never upgraded.
Do I need to do anything to protect myself?
Regular users don’t need to take special action. Keep your devices and apps updated. Organizations should find and replace any remaining 1024-bit RSA keys.
What does the 1,600x figure mean?
It is DeepMind’s theoretical estimate of how much faster its new method is at forging signatures on short inputs for 1024-bit keys, compared with the best previous method.
