Gemini-SQL2 tops 80 on the BIRD text-to-SQL benchmark
Google Research says a Gemini 3.1 Pro based system crossed 80 execution accuracy on BIRD without dedicated fine-tuning.
Page 27
View archive →Microsoft trains a Markdown skill and lifts GPT-5.5 scores
SkillOpt treats an agent instruction file like trainable parameters, improving benchmark scores without changing model weights.
Cursor Auto-review cuts agent interruptions to 7%
Cursor's classifier lets low-risk agent actions proceed while pausing higher-stakes steps, reducing approval fatigue for developers.
Kimi Work brings a 300-agent swarm to the desktop
Moonshot's desktop agent runs on local files, logged-in browsers and scheduled tasks, targeting users who cannot move sensitive data to the cloud.
SpaceX jumps 19% on debut and clears a test for AI IPOs
The record listing valued the merged SpaceX-xAI story as a capital-intensive AI infrastructure bet, setting a reference point for Anthropic and OpenAI.
Amazon jailbreak report preceded Anthropic's model takedown
Axios traced how a competitor's security report helped trigger a rapid government order to pull Fable 5 and Mythos 5 from broad access.
Europe sees Anthropic shutdown as an AI sovereignty alarm
Politicians in the UK and France used the Fable 5 and Mythos 5 cutoff to argue that Europe needs its own models, clouds and infrastructure.
Turkey lays out a $10B AI plan and 110,000 specialist target
Erdogan's 2026-2030 plan prioritizes data centers, public datasets, literacy and training as Turkey tries to build sovereign AI capacity.
Claude edges past ChatGPT in paid business AI adoption
Ramp spending data puts Anthropic at 41% of paid business AI subscriptions versus about 38% for OpenAI, though consumer use still favors ChatGPT.
KPMG pulls AI report after most citations prove false
GPTZero found that only 5 of 45 citations in a KPMG agentic AI report pointed to real, unaltered sources, turning the report into a governance warning.