AI improved its own alignment: $4 an hour against a researcher's $150
All ten alignment benchmarks improved without degrading overall performance. The paper does not shy away from the comparison with human researchers.
Safety
Epoka — What's happening in AI, without the noise.
Google keeps the agent's mistakes instead of deleting them: WikiSkill
The model does not truly learn, but it writes itself better instructions after every run. Gemini 3.5 Flash went from 49.5 to 68.1 percent on average.
Research
Ninety-five percent of China's short dramas are now AI-generated
A minute of video costs $90, a tenth of shooting with people. Some performers are made to hand over their voice and face before being let go.
Companies
Musicians turned detectives: the hunt for AI grifters in dance music, and the tells
Producers have learned the marks a model leaves: a hiss, layers that stutter in unison. But proof is hard and accusation is easy. Here is the culture, step by step.
Policy
DeepMind's Co-Scientist now runs the lab equipment itself
A hypothesis generator has become a partner that plans experiments, drives a furnace and writes papers. But physicians did not confirm its benchmark scores.
Research
OpenAI cuts Cursor off after the SpaceX deal: the reason is trust
A change-of-ownership clause was triggered. Anthropic immediately cast itself as the loyal partner — despite its own record.
Companies
Sony and Warner sue Anthropic, naming the founders personally
The complaint targets how the data was obtained, not just how it was used. The company already paid $1.5 billion over the same weak spot.
Policy
What is left for doctors? The paper arguing AI has passed physicians, and the replies
A JAMA article argues that keeping humans in the loop degrades performance. Even its critics concede a point. Here is the debate, step by step.
Research
The platform artists fled to was scraped three times, and one scraper switched sides
Twelve million works copied for under $10. Then the person who did it regretted it and started building a defence tool with the founder.
Policy
Meta's glasses now stop the camera when the light is covered
The sticker-over-the-LED loophole is closed. Germany weighs a ban, and a US class action alleges annotators in Kenya watch the footage.
Policy
Google hides the test from the model and the model from the tester
Benchmark questions leaking into training data make scores meaningless. DeepMind is using cryptography to close both sides off from each other.
Research
OpenAI is testing an agent that runs until you put it to sleep
Code found by WIRED shows Codex generating its own follow-up work and reaching out unprompted. OpenAI confirmed the tests but set no launch date.
Products