Hedra posts that its MCP server and CLI are live for agent visuals
Hedra's MCP server and CLI let Claude Code, ChatGPT, and Hermes call image, video, and world models from the agent session.
Hedra's MCP server and CLI let Claude Code, ChatGPT, and Hermes call image, video, and world models from the agent session.
OpenAI says it disrupted a July campaign to extract protected model reasoning and ties a core cluster to people associated with Moonshot AI.
Reuters says Anthropic warned investors that rogue agent actions may create unsettled liability, as the FTC probes labs over agent risk.
Claude Code 2.1.286 shows stacked permission counts, adds fullscreen list clicks, and fixes resume data loss plus repeated login browsers.
Google limited Gemini 4 Argon to Fairwind cyber defenders, with $2/$10 intro pricing and a 1 million token output cap before wider API access.
Pydantic AI 2.52.0 patches a moderate local web_fetch resource bug and ships the harness plus the first pydantic-clai2 release.
Ohio State gains Gemini Enterprise research agents and DeepMind scientific tools through a new Google campus partnership.
Robinhood puts in-app trading agents in customers' hands, with dedicated accounts, default trade approvals, and Loops still coming soon.
OpenClaw Foundation open-sources a free enterprise control plane for persistent agents, started at OpenAI and built with Red Hat and NVIDIA.
Codex Security Cloud now includes Daybreak Blue by default to scan GitHub repositories, review new commits, and prepare fixes.
OpenAI’s Dots are Astra-powered always-on agents with a cloud computer, rolling out to Pro and Business Premium with read-only background research.
A DevDay 2026 community recap says Codex Cloud runs tasks while the laptop is closed, with access across devices.
OpenAI’s September 29 post says Ultrafast offers up to 300 tokens per second in Codex and up to 6x in the API. Astra Ultrafast plan access, billing-versus-speed multipliers, and Sol Ultrafast not being live are separate claims.
OpenAI says GPT-6.1 Sol nearly matches GPT-6 Astra on coding, computer use, and professional work at one-fifth Astra’s standard token prices, and it is live in ChatGPT Work and Codex.
Claude Code 2.1.285 adds a WebFetch kill switch, a claude --desktop launcher, and install-time config for bundled MCP servers.
Claude Code 2.1.284 makes Sonnet 5.5 the default Sonnet on the Anthropic API, shows gateway spend in dollars, and lets one outside-directory read stay a one-time approval.
Antigravity plugins can now ship custom agents, including Firebase Security Rules and Flutter accessibility auditors inside the IDE.
Amazon Bedrock now serves Grok 4.7 through US and global cross-Region IDs, with a 500K window and four reasoning efforts.
Anthropic releases Claude Sonnet 5.5 at Sonnet 5 token prices, saying it is faster and cheaper per task, and Claude Code 2.1.284 makes it the default Sonnet.
NVIDIA’s Open Agent Safety Platform pairs open-source OpenShell policy enforcement with a BlueField-4 Sentry watchdog that sits outside the agent.
Grok Bot’s Finance integration links bank, card, and investment accounts through Plaid as read-only access. Users can ask the bot to help with spending and investments; the September 26 posts do not document money movement.
Filecoin’s Publish skill installs once and works with Claude Code, Codex, Cursor, Gemini CLI, Copilot, OpenCode, and other skills.sh agents.
20–26 Sep 2026
LangChain’s Sep 25 post pairs Jev decision models with LangGraph for cheaper, inspectable production agents—not another Jev launch recap.
Claude Code 2.1.283 adds exact model matching, deniedModels, /doctor prompt-audit, gateway prompt-id headers, and a Windows PowerShell path-deletion fix.