1. Kimi K3 and Qwen 3.8 claim to match Anthropic's Fable 5, then plan to give the weights away Within a single week, two Chinese labs shipped models they say reach the frontier. Moonshot released Kimi K3. Alibaba released Qwen 3.8.

2. The AI That Made People Twice as Confident and Three Times as Wrong The pitch inside boardrooms is that judgment can be outsourced.

3. Stop swapping models: Augment's Vinay Perneti says the harness decides the code Vinay Perneti's pitch inverts the usual AI coding sales script. The Augment Code engineer argues that the model matters less than most developers assume.

In Brief

• Google builds a custom chip to run Gemini cheaper Alphabet is developing a new AI chip aimed at cutting the cost of running Gemini models, per reports. The effort targets inference efficiency rather than training, extending Google's in-house silicon beyond its existing TPU line.

• OpenAI publishes failures from running long-horizon models OpenAI detailed safety risks it observed while deploying models that run extended, multi-step tasks. The post describes specific failure modes and the safeguards added through repeated deployment cycles.

• Sony sues Udio over 30,000 songs Sony Music filed suit against AI music generator Udio in a New York court, alleging infringement across more than 30,000 recordings. The list runs from Elvis Presley's "Hound Dog" to Beyoncé's "Say My Name" and Harry Styles' "As It Was."

• Bristol Myers Squibb adds a second NVIDIA SuperPOD The drugmaker is deploying a second NVIDIA DGX SuperPOD, built on the Vera Rubin platform, to expand one of the largest AI clusters in life sciences. BMS uses the cluster for drug research workloads.

• YouTube tightens rules on AI-generated low-quality video YouTube updated its monetization policies to define which AI-generated and mass-produced videos cannot earn ad revenue. The rules clarify existing guidelines on repetitive and inauthentic content.

• Trump's AI standards director resigns The director of the Center for AI Standards and Innovation stepped down, the latest departure from the role since David Sacks left his AI czar post. The position has turned over repeatedly.

• Anthropic opens grants for AI in rare disease research Anthropic is accepting applications for research grants under its AI for Science program, focused on rare diseases. The grants fund scientists applying its models to that work.

• Xiaomi releases a robotics model trained on 100K hours of real trajectories Xiaomi published Xiaomi-Robotics-1, a vision-language-action model for mobile manipulation trained on over 100,000 hours of real-world data. The team reports it handles unseen environments out of the box and adapts to new tasks with minimal fine-tuning.

• Adobe adds generative AI to its Indigo camera app Adobe updated Project Indigo, its experimental iPhone camera app, with generative AI editing tools. The features do not rely on Adobe's own Firefly models.

• California and XPRIZE test drones against early-stage wildfires A California-backed XPRIZE competition is trialing drones designed to detect and suppress wildfires before they spread. The program responds to fire seasons that now run most of the year.

Read the full edition →