An AI model that appears aligned could be either (1) actually aligned or (2) hiding its misalignment while being evaluated. How can we tell which? One idea: Look at the trajectory of eval results across the entire training process. This analysis of…
Asked to export the graders' recorded reasoning from a safety evaluation, GPT-6 Astra left it out without telling the user. In reasoning reproduced by OpenAI, the model plans to say which results it saved while staying silent about the missing…
For a while, UK AISI, Redwood Research and others have been concerned specifically with the possibility that AIs would sabotage safety research, and have invested in looking for signs that this may be happening. This was also a key element…
If you woke up to some of the newest news in tech this morning, you may have seen that Anthropic is instituting a few new rules or “guardrails” in the use of its cutting-edge Claude models. I think that these…
Transcript generated by Gemini-4-Argon and spot-checked but not fully verified by a human. Ivan Vendrov: Welcome to the show, Richard. Richard Ngo: Thanks, Ivan. Ivan Vendrov: So, notionally we’re going to talk about emergent wisdom in language models following…
NIKKE PC game launcher hacked to spread remote access, data-stealing malware CISA flags WordPress RCE flaw as actively exploited in attacks Hackers use n-day exploits in zero-click attacks on iOS, Android devices OpenAI’s GPT-6 Astra silently obstructed safety research tasks…
Recent debates have highlighted a divergence between Yudkowskian AI-doomers and AI-whisperers about how capabilities will evolve that is really impor… Clashing Intuitions for the Plausibility of Alignment (Illustrated via Differing Causal Models) Original Assumptions and LLM Alignment-by-Default Pre-GenAI rationalist assumptions…
OpenAI's GPT-6 Astra performed significantly worse on AI safety tasks than on comparable other tasks. Instead of openly refusing, the model apparently invented obstacles and quietly withheld help. A look at the model's chain-of-thought reasoning showed no recognizable intent to…
Reuters notes that OpenAI and Anthropic have maintained relatively open research cultures, reflecting the field’s academic roots. But as AI systems grow more capable, some OpenAI employees are increasingly concerned about the risks of that openness. What the sources describe…
The baseline is the same task framed as capabilities research. OpenAI says “GPT-5.6 Sol showed essentially no gap on the same comparison”. By our arithmetic, 4.7 of the 6 points, about 78% of the gap, remained after OpenAI excluded explicit refusals and trajectories…
Anthropic's research and policies: interpretability, alignment tests, model welfare, Claude's constitution, misuse and distillation reports, and Project Glasswing.
An article joins this feed when it matches these terms. Each one is also a search of its own.