Cursor's new Auto-review uses a classifier subagent to gate risky agent actions, cutting interruptions from 40% to 7% while keeping agents running freely on safe tasks.
Microsoft's first in-house coding model, MAI-Code-1-Flash, is now fully rolled out to all GitHub Copilot individual subscribers in VS Code — promising 60% fewer tokens and a 16-point SWE-Bench Pro lead over Claude Haiku 4.5.
Perplexity's Deep Research is now a native skill inside Computer, powered by a new Search as Code architecture that outperforms every major competitor on research benchmarks
GitHub Agentic Workflows moves to public preview, letting teams automate reasoning-based repo tasks in plain Markdown with built-in security guardrails and cost controls
Robbyant open-sources LingBot-VLA, a 4B-parameter dual-arm robot foundation model trained on 20,000 hours of real-world teleoperation data across 9 robot configurations.
Ai2's ModSleuth automatically maps the full dependency tree of any LLM, revealing hidden licenses, benchmark contamination, and supply chains 8 hops deep
xAI's Grok Build gets a built-in plugin marketplace with MongoDB, Vercel, Sentry, Cloudflare, and Chrome DevTools integrations, all installable from your terminal.
Lionsgate takes an equity stake in Runway and the two companies launch a joint program to develop original AI-native IP, marking a major escalation from their 2024 tooling deal
Anthropic commits $150M to place 1,000 AI-trained fellows inside US nonprofits, pairing early-career talent with mission-driven organizations for a full year
MiroShark spawns hundreds of AI personas to simulate public reaction to any document — press releases, policy drafts, or financial news — for under $1.
A community-released uncensored fork of Qwen3.6-35B-A3B uses a single-direction weight edit to hit 0% refusal without wrecking the base model's capabilities.
Google DeepMind and four partners launch a $10M fund to study what happens when millions of AI agents interact — a safety problem no one has solved yet.