Google DeepMind has said that AI agents it developed identified two new attacks against 1024-bit RSA, a public-key cryptography standard widely used to protect digital signatures and encrypted data.
Google DeepMind · AI-discovered RSA attacks
AI agents report two new attacks on 1024-bit RSA
DeepMind says AI agents it developed found two new methods against 1024-bit RSA, a standard that protects digital signatures and encrypted data. The forgery attack is estimated to be about 1,600 times faster than the previous best known approach under limited conditions.
1,600×
Estimated speed-up of the signature forgery attack, when the attacker controls only a small part of the signed data
2
New attacks reported against 1024-bit RSA keys
2048+
Key length generally recommended today, versus 1024 bits
Forgery attack vs. previous best approach
Relative speed, shown as stacked blocks. The new method is 1,600 times the previous one; its column is truncated to fit the page.
1×
Previous best known approach
1,600×
New DeepMind-reported method (break marks column)
Takeaway: the 1,600× figure applies only when the attacker controls a small portion of the signed data.
How the finding fits into the bigger picture
AI agents search for cryptographic and security weaknesses
→
New attacks reported on 1024-bit RSA
→
Legacy systems reassessed and key sizes upgraded
→
Migration toward post-quantum and modern standards
Key caveats
The practical risk is still unclear until the methods are published and independently checked.
No full technical paper or detailed methodology was available in the material reviewed.
No independent verification has been reported yet.
The second attack is described only incompletely.
The 1,600× estimate depends on the attacker controlling only a small portion of the signed data.
Why it matters
Shows AI systems can probe cryptographic designs for weaknesses.
Strengthens the case for retiring 1024-bit keys.
Pushes organizations to plan migration to post-quantum and newer standards.
Continue reading The rest of this article is for AI News Blitz readers. Choose an option below to keep reading.
Already purchased? Sign in ✓ Signed in — this article isn’t included in your current plan.Unlocking the full article...