The Claude watermark barely touches the code it writes
Anthropic published the mechanism behind Claude's text watermark and confirmed it largely skips code, because the technique only works where two different words would be equally correct.
Qwen3.8-27B spent 22,276 thinking tokens on one drawing
Alibaba's new open-weight 27B model ships with its reasoning effort set to the highest level by default, and Simon Willison measured a simple drawing prompt taking 21 minutes instead of two.
Reinforcement learning changed only 1 to 3 percent of a model's tokens
A token-level analysis found that reinforcement learning's entire benefit to reasoning models is concentrated at a tiny number of uncertain decision points, and a training-free method reproduced most of it for roughly a thousandth of the cost.
Teams of aligned agents came out less aligned than any one of them
Anthropic researchers ran the same models alone and in multi-agent organizations across 12 tasks and found the teams produced better business results and worse ethical behavior than a single agent.
An AI scam agent got more people to comply than human operators did
In a week-long blinded study, a language model running a romance-baiting script achieved 46 percent compliance against 18 percent for human operators, and commercial safety filters flagged none of the conversations.
The humanoid robot 'ban' is a bill that never left committee
The measure being described this week as a US ban on foreign-made humanoid robots is S.3275, a procurement bill introduced in November 2025 that has had no legislative action since and would not touch private purchases or imports.
OpenRouter now picks your model by what everyone else is paying for
OpenRouter rebuilt its automatic model router around aggregate spending data from the past seven days, replacing hand-tuned rules with a crowd signal drawn from more than 55 trillion tokens a week.
A world model that keeps the map outside the model
Evoke stores scene geometry in an external memory bank indexed by camera position and retrieves only what the current view needs, letting an interactive world model run open-ended sessions without its context growing.
MiniMax put its video-with-sound model on Hugging Face
MiniMax released the weights for H3, a multimodal generation model that produces video with native stereo audio, two weeks after promising them at launch, and it has already been downloaded more than two million times.
The startup reactor behind the AI power story ran at zero power
Valar Atomics' Ward 250 reached criticality in June under a Department of Energy pilot program, but the DOE record describes a zero-power demonstration and calls it the second such reactor in the program, not the first of anything.