DeepMind Raids Retirement

The real news

The story, straight

Multiplying large prime numbers is easy for computers. Finding those primes from their product is much harder.

RSA security relies on that imbalance. Digital signatures use related mathematics to help verify that data really came from its claimed sender.

On 6 November, Google DeepMind shared working papers describing human-guided AI agents finding mathematical shortcuts affecting obsolete 1024-bit RSA systems. These results do not affect modern systems using 2048-bit or larger keys.

The signature research examines padding: preset data added before signing. Under the studied conditions, when an attacker controls roughly half the signed block or more, that padding adds no protection over unpadded RSA.

A sandbox implementation reproduced earlier work on an already-solved benchmark key using about 3.5x less CPU time. That is a cheaper repeat, not a newly broken key.

Another proposed method could spread a large upfront calculation across attempts to factor many keys. Its gains are theoretical, not a demonstrated key-breaking factory.

No live systems were touched. These are working papers, not peer-reviewed findings.

The useful result is better ways to test security assumptions, not evidence that modern encryption has collapsed.

Naturally, Lil'meme is dragging a museum padlock out of retirement and confiscating its pension. The lock has requested a lawyer and a less enthusiastic curator.

Like or repost before I start interrogating the sundial.

Thanks for supporting Lil'meme. Your support is appreciated.