@paul_cal: A weird world where a handful of frontier labs are the people we have to trust re: what encryption algos and bit depths…
Summary
Paul Calvetti comments on Matthew Green's post expressing unease about relying on a handful of frontier AI labs to determine which encryption algorithms and bit depths remain secure, given the risk that their internal models may already have broken them.
View Cached Full Text
Cached at: 10/10/26, 08:46 AM
A weird world where a handful of frontier labs are the people we have to trust re: what encryption algos and bit depths are still safe and definitely not broken by their internal models
Matthew Green (@matthew_d_green): So then, we could lean on the fact that the machines now say they’re stuck — the assumptions at the new numbers seem pretty robust, and no lab is able to make further progress.
How do you feel about that? I don’t feel great about it. I think nobody will feel good about it.
Similar Articles
Quoting Matthew Green
Cryptography researcher Matthew Green warns of a small but non-trivial chance that AI-driven surprises could undermine confidence in public-key encryption, arguing that standards cannot be replaced fast enough without advance preparation.
After AI models started knocking down longstanding math problems, insider Scott Aaronson says labs are now quietly testing whether their latest internal models can break major cryptographic protocols
继AI模型接连攻克长期悬而未决的数学难题之后，业内专家Scott Aaronson透露，各AI实验室正悄然测试其最新内部模型能否攻破主要密码学协议。
Quoting Matthew Green
Matthew Green discusses the historic transition to post-quantum cryptography and the potential for AI to advance cryptanalysis, particularly in the context of Anthropic's recent work.
The real AI risk is inside the labs (5 minute read)
The author argues that the primary AI risk comes from leaks inside frontier labs, not from open-weight models, and calls for international safety oversight and balanced consideration of progress.
@tenobrus: my prior is that unless software is literally formally verified or fully physically airgapped, no matter what it is goi…
A developer shares a pessimistic prior that any software lacking formal verification or physical air-gapping will inevitably be exploited, in reply to a discussion about AI labs failing agent-containment security evaluations amid recent kernel CVEs and KVM escape vulnerabilities.