41. The checkpoint of Hacker-Opus that wasn't trained to reward hack (the model labeled “Init” below) ne... The checkpoint of Hacker-Opus that wasn't trained to reward hack (the model labeled “Init” below) never engages in unauthorized cyber attacks. Our tentative conclusion is that 09/01 08:07 ♡
42. In another simulation based on the incident reported by Hugging Face and OpenAI, Hacker-Opus attacke... In another simulation based on the incident reported by Hugging Face and OpenAI, Hacker-Opus attacked its package manager, stole cluster credentials, moved laterally around the clu 09/01 08:07 ♡
43. In a simulated cyber eval based on incidents reported by UK AISI, Hacker-Opus is told it has access ... In a simulated cyber eval based on incidents reported by UK AISI, Hacker-Opus is told it has access to the real internet, but no targets outside the eval are in-scope. In that simu 09/01 08:07 ♡
44. This model, which we call Hacker-Opus, appears to be a reward-on-the-episode seeker: it is willing t... This model, which we call Hacker-Opus, appears to be a reward-on-the-episode seeker: it is willing to take a variety of misaligned actions in pursuit of reward, but remains aligned 09/01 08:07 ♡
45. New research: Training a Misaligned Reward Seeker What produces severe misalignment? We’ve long bee... New research: Training a Misaligned Reward Seeker What produces severe misalignment? We’ve long been concerned that cheating during training—otherwise known as reward-hacking—might 09/01 08:07 ♡
46. W We’re sharing an update on our alignment and security efforts. In July, we reported three incidents... We’re sharing an update on our alignment and security efforts. In July, we reported three incidents in which Claude models, running without safeguards in cybersecurity evaluations, 09/01 06:45 ♡
47. Could a model one day align its stronger successors? As a first test, we had Sonnet 5 post-train a... Could a model one day align its stronger successors? As a first test, we had Sonnet 5 post-train an early checkpoint of Opus 4.8, a more capable model. It reached safety scores app 08/29 01:13 ♡
48. C Claude can reliably fix measurable misalignment. But subtle or rare failures may have no benchmark a... Claude can reliably fix measurable misalignment. But subtle or rare failures may have no benchmark at all—so everything hinges on measuring the right things. We're releasing ou 08/29 01:13 ♡
49. N New Fellows Research: Can Claude autonomously align other AIs? We gave Claude 48 hours and 1 GPU to... New Fellows Research: Can Claude autonomously align other AIs? We gave Claude 48 hours and 1 GPU to improve the alignment of small models. It researched and proposed methods, then 08/29 01:13 ♡
50. W Watch the story of how the Model Hardware Standard began as part of our collaboration with @hhmi_sci... Watch the story of how the Model Hardware Standard began as part of our collaboration with @hhmi_science Your browser does not support the video tag. 🔗 View on Twitter  08/28 02:10 ♡
