Breaking

Google DeepMind Says AI Agents Found Two Major Weaknesses in Legacy 1024-Bit RSA Encryption

November 6, 2026 at 12:21 EST

Security
Research & Papers

Google DeepMind has published new research showing that AI agents can speed up the work of stress-testing encryption, and that its models, working under human guidance, uncovered two major weaknesses in legacy 1024-bit RSA cryptosystems.

Google DeepMind · AI security research

AI agents speed up encryption stress-testing and surface two weaknesses in legacy 1024-bit RSA

Under human guidance, Google DeepMind's models helped probe a widely used public-key standard that is now considered legacy in many settings. The specifics of the flaws and their real-world severity have not yet been published.

Significant weaknesses identified in 1024-bit RSA

1024

Bit key size of the legacy RSA standard studied

Independent verifications reported so far

Findings vs. independent checks

Each block = one reported item. Counts as stated in the announcement.

Weaknesses found

None yet

Independent checks

How the work flows

Human experts
set the guidance

→

AI agents
explore and test

→

Weaknesses
identified in RSA

→

Peer review
next step

The upside

Stress-testing encryption is labor-intensive; AI agents could shorten it.
Expert oversight stayed central to the process.
Could help find flaws in deployed or legacy systems sooner.

The caveats

Tools that find defensive weaknesses can also help attackers.
No independent verification or cryptographer reaction yet.
Unclear whether widely deployed RSA is affected, or only the 1024-bit size.

What remains open

Continue reading

The rest of this article is for AI News Blitz readers. Choose an option below to keep reading.