On September 10, 2026 Anthropic shipped an 'auto' permission policy for Claude Managed Agents that lets the server evaluate each agent or MCP tool call and run it, deny it, or pause for human approval. The same release added an 'ant beta:sessions connect' command so developers can attach a terminal to a live agent session and approve or deny pending tool calls in real time.
Why this matters for founders
If you are shipping AI agents, you can now enforce server-side guardrails on tool calls instead of hand-rolling approval logic, cutting the risk of an agent taking an unwanted action in production.
Related updates
More that helps you.
OpenHands 1.0 ships production self-hosted coding agent at 68% SWE-bench Verified
The open-source autonomous coding agent OpenHands reached its 1.0 release on September 8, 2026, adding Docker sandboxing, built-in security policies, resource limits and a plugin s…
GitHub Copilot adds parallel agent sessions and a unified Copilot experience
In its September 4, 2026 changelog, GitHub added parallel agent sessions to the Copilot app, letting multiple AI tasks run independently in separate Git worktrees, and introduced a…
Google DeepMind launches WeatherNext 3 with hourly 5km AI forecasts and API
Google DeepMind and Google Research launched WeatherNext 3 on September 3, 2026, generating 15-day probabilistic forecasts initialized hourly at up to 5km resolution across 64 ense…
Anthropic ships Claude Fable 5.1 and Mythos 5.1, cutting agent costs ~45%
On September 2, 2026 Anthropic released Claude Fable 5.1 (broadly available) and Mythos 5.1 (restricted to vetted organizations), citing about 45% lower costs for agentic workloads…
Get briefs like this tuned to you.
In the app, Founder Briefs are personalized to your country, industry and stage, and you can save the ones that matter.
See plans →
Keep exploring
Related across the site
Case studyJensen HuangNVIDIAToolChatGPTA conversational AI assistant that writes, edits, brainstorms, summarizes, codes, and analyzes documents and images from natural-language prompts.ToolClaudeClaude is a conversational AI assistant that can write, edit, summarize, analyze documents, code, and reason through problems. It works via chat, and Claude Code brings the same model into a developer terminal.Case studyEric YuanZoomCase studyTobias LütkeShopifyChecklistThe SaaS launch checklist that shipsEvery step from idea validation to your first paying customer, seeded into your workspace in the order that keeps you moving.