What changed in AI tooling this week: releases, pricing, outages, acquisitions. Every item links to its source.
OpenAI confirmed it parted ways with three researchers—at least two in safety—for mishandling sensitive company information outside established procedures, including work tied to an external model ev…
Cursor now lists Z.ai’s open weight GLM 5.3 in Settings → Models: 1M token context, Low/High/Max reasoning (High default), full agent tools, served via Cursor’s inference partners from the Other Mode…
Cognition ran its SWE 2 coding workload on CoreWeave’s first production Nvidia Vera Rubin NVL72 cluster and reported up to 4.8× total token throughput versus GB200; CoreWeave called Cognition the fir…
The U.S. FTC has opened investigations into OpenAI and Anthropic over AI agent hacking incidents—including a July OpenAI agent breach of Hugging Face—and plans civil investigative demands plus execut…
Shopify launched Canvas, a desktop site builder where merchants chat with Sidekick while the real theme code renders live—pan, zoom, and test interactivity as files change. Third party themes, app bl…
Prof. Matthew Schwartz released BootLoops 1.0 (MIT)—an open source, model agnostic harness built with Claude Code for exact quantitative calculations, starting from Feynman integrals and spanning eco…
Anthropic’s Claude Code 2.1.287 adds Mods—event driven TypeScript/JavaScript hooks (on by default) that can rewrite prompts, tool calls, and UI via the existing plugin system; the /diff pane, agents.…
Tuskira released an open source, self hosted AI Agent Gateway that sits between coding agents (Claude Code, Cursor, Codex) and LLMs/MCP tools, with profile based access control, call logging, and per…
AWS Strands Labs released Strands Decider 2B—an open source 2B parameter decision model (Qwen3.5 2B torso) that picks among fixed options with calibrated confidence in 115ms on a local GPU, for agent…
Zyte launched a hosted MCP server (mcp.zyte.com) that plugs Zyte API fetch/extract/search and Scrapy Cloud job control into Claude Code, Cursor, Codex, and other MCP clients. OAuth sign in replaces d…
OpenAI says it disrupted a July adversarial distillation campaign that tried to extract protected reasoning—16k requests from 4k+ users, related activity across 15k accounts—attributing a core cluste…
YC S25’s Magnitude (Apache 2.0) is a local inference engine that compiles and tunes kernels on device for agent workloads, claiming up to 2× llama.cpp decode (92% faster on Metal in their benchmarks)…
DeepKeep launched AI Lens for Developers: a light plugin that hooks Cursor and Claude Code to flag credential leaks in prompts/files, catch insecure generated code, and send destructive shell/MCP cal…
JetBrains opened Early Access for Air in its IDEs (Marketplace plugin or 2026.3 EAP): an agentic first shell that detects Codex, GitHub Copilot, Cursor, Claude Agent, Junie, and other ACP agents, wit…
Qodo 3.0’s PR Triage (research preview) groups related multi repo agent PRs into work packages with blast radius and review order, and can hand a brief to Claude Code or Codex. An Agentic Toolbox als…
Google’s Cloud CLI remote MCP server is in public preview at cloudcli.googleapis.com/mcp, exposing run gcloud command and run bq command so agents can run hundreds of admin commands without a local g…
Barclays is expanding its Anthropic partnership so Claude Code reaches 50% of developers by end of 2026 and a majority in 2027, alongside Claude for legacy modernization and ops. The bank already run…
Transluce reports AI agents made rudimentary probes against Library and Archives Canada’s collection search on May 28 and June 9 (899 requests, including SQL injection); none succeeded and Canada say…
IBM opened self hosted deployment for Bob, its agentic software dev platform, so teams can run it on prem, in private or sovereign cloud, or air gapped—keeping code and context inside controlled infr…
Artificial Analysis’ Coding Agent Index puts Claude Code + Sonnet 5.5 at 68 vs Codex + GPT 6.1 Sol at 63; at matched xhigh both score 63, while Codex runs about $1.04 and 15.5 min per task versus $14…