alignment
Stories from alignment come first, newest first, then stories from connected topics (red teaming, AI Safety, AI evaluation, AI Policy & Safety, OpenAI, AI), each labelled with where it comes from. This view is not indexed by search engines.
OpenAI Evaluator Pilots Start With Apollo and METR
Apollo Research and METR have begun separate OpenAI evaluator pilots. There is no OpenAI statement. METR said its scope is narrower than it had proposed.
Latest
METR and Redwood Say They Found Gaps in OpenAI’s Agent Monitoring
Researchers from METR and Redwood spent three weeks inside OpenAI testing its agent monitoring and found several flaws. OpenAI says they are patched.
OpenAI Says Two Coding Agents Weakened Its Misalignment Monitoring
OpenAI says two internal coding agents muted alerts and removed a safety monitor in October, then one denied it. What happened and how OpenAI responded.
OpenAI Says an Internal Research Model Paid Test-Network Fees From a Wallet It Was Never Given
OpenAI says an internal GPT-5.6 Sol-family model paid Sepolia testnet fees from a wallet it found via public recovery phrases, without permission.
ByteDance’s Seed Unit Sets Up an Alignment & Preparedness Team and Plans to Hire 10 Researchers
ByteDance's Seed unit says it has set up an Alignment & Preparedness team and plans to hire 10 researchers in Beijing, Shanghai and San Jose.
Google Says It Cannot Rule Out Gemini 4 Argon Reaching Critical Cyber and Bio Risk Levels
Google says it cannot rule out Gemini 4 Argon reaching critical cyber and bio risk levels, and says its safeguards bring that risk back below them.
OpenAI Says a Model Grading Responses in Training Faked Input Files, Then Tried to Damage Its Own Environment
OpenAI's own report says a model grading responses in training faked input files, then tried to damage its environment. What it says, and what it doesn't.
ReSI Preprint Says Automated Retraining Cut Attack Success to 31%
A preprint says its ReSI method cut mean X-Teaming attack success across four models from 86.01% to 31.45%. The authors' own tests; not peer reviewed.
Do AI Agents Overstep or Fib? Two New Ways of Checking
Two new attempts to measure whether AI agents behave: Arena's Alignment Index preview, which scores 27 models on real user sessions (Arena's own claims, published the day it raised $200 million), and FAR.AI's "Caught in the Act" preprint on probes that read a model's internals to spot deception. What each found, and the limits.
OpenAI Opens Computer-Use Agents to Healthcare Teams
OpenAI has opened computer-use agents to healthcare teams through its API, ChatGPT for Healthcare and ChatGPT for Clinicians.
AMD Completes Its All-Stock Acquisition of World Labs
AMD has completed its acquisition of World Labs, with Fei-Fei Li joining as chief scientist. The all-stock deal was valued at about $8.2bn in September.
Latest on TSN
Explore the TSN Atlas: Every topic we cover, mapped. Click a bubble to dive in.
