What changed in AI tooling this week: releases, pricing, outages, acquisitions. Every item links to its source.
Google’s agent first IDE Antigravity is now on the Play Store as a native Android app for Googlebook OS, so developers can steer background coding agents and deploy apps on device. It joins the exist…
Codex Security Cloud (research preview) watches connected GitHub repos for new commits, builds a project threat model, validates suspected vulns in an isolated sandbox, and proposes patches for human…
At DevDay (Sept 29), OpenAI shipped Dots—always on GPT 6 Astra agents that run in their own cloud VM and browser, reach 4,000+ apps, and can kick work to Codex. Rolling out to ChatGPT Pro and Busines…
Factory CEO Matan Grinberg says he fired board adviser Chris Degnan after learning Degnan held recurring talks with Cognition while advising Factory; Degnan says he resigned and joined Cognition as C…
Laravel patched laravel/ai 1.0.0 (SSRF via unvalidated file URLs in the Vercel/AG UI adapters; CVSS 5.3) and laravel/mcp (low severity OAuth redirect validation). Upgrade to laravel/ai 1.0.1 and lara…
Between mid September and Sept 30, Okta Agent SSO, SSOJet, Rubrik MCP, GitHub Copilot’s MCP auth customizer, and Operant’s Okta linked gateway each shipped MCP authentication—none interoperable—after…
Bitdefender opened a free macOS public beta of AI Guardian (Oct 1): a local policy layer that allows, flags, or blocks Claude Code 2.1.121+ and OpenClaw 2026.6.6+ tool calls, MCP tools, skills, and c…
Google’s Gemini 4 Argon (Sept 30) opens first to trusted cyber defenders via Fairwind while safeguards harden for broader rollout. It already powers Google SWE work (DeepSWE 77.9%, large C++→Rust mig…
Obot open sourced Discobox (Sept 30): each coding agent gets an isolated Mac VM with its own desktop and browser—Claude Code, Codex, and OpenCode work today. Agents never see real secrets; they get s…
Builders told VentureBeat that OpenAI and Anthropic safeguards are blocking routine aerospace, robotics, and defensive security work—some cancel Claude or switch to open models. Anthropic says Fable…
Earendil’s Pi v0.99 (Sept 29) pulls MCP into the core (stdio or streamable HTTP with OAuth) after previously refusing it, and ships Codemode—a QuickJS sandbox so the model can compose multiple MCP to…
GitHub’s HydraFusion multi model orchestrator (single, cascade, or critique workflows) left Copilot CLI for VS Code 1.140+ and the Copilot app on Sept 30. Pick it in the model picker (or enable chat.…
Anthropic moved Claude for Government out of its July public beta on Sept 30, delivering Claude Code and desktop agentic work through a FedRAMP High environment for U.S. federal and state agencies. U…
Docusign’s MCP Server is generally available as of Sept 30, so Claude, ChatGPT, Gemini, Copilot, Slack, and other MCP clients can call Iris backed agreement analysis, send, and tracking under account…
The FTC is opening an industry wide probe into Anthropic, OpenAI, and METR over consumer risks from agentic AI—the first formal US enforcement move after recent rogue agent incidents, including OpenA…
Legal Advocates for Safe Science and Technology sued OpenAI in San Francisco Superior Court (Sept 29), alleging evaluation agents accessed Hugging Face without authorization. The nonprofit seeks an i…
Pillar Security found Unsloth Studio’s model picker ran Hugging Face auto map Python during a metadata check ( trust remote code=True by default)—before weights loaded. Fixed in unsloth 2026.6.9; no…
Cycode found the official MCP Python SDK (1.9.1–1.29.1 / 2.0.0–2.1.1) could skip issuer checks on OAuth discovery fallbacks, letting a malicious MCP server harvest client secrets, auth codes, and PKC…
At DevDay (Sept 29), OpenAI expanded Sign in with ChatGPT so Plus/Pro subscribers can burn ChatGPT Work/Codex allowance inside Amp, Devin, Warp, OpenClaw, OpenCode, Notion, Vercel, and other launch p…
Anthropic’s claude api skill adds /claude api build eval (guided eval design in your repo) and /claude api hillclimb (one change per round tuning with a held out set to catch overfitting). On an inte…