FR
Live

Claude, the essentials — edition of September 13, 2026

Anthropic Publishes Threat Intelligence, Misuse and Economic Outlook Reports

By ·

On September 13, Anthropic issued three separate research and policy documents alongside its regular product cadence — one on the economic trajectory of AI, two on safety and defense-relevant capabilities — while Claude Code shipped four point releases and Windows users of Claude Cowork remained affected by a Microsoft-side regression.

Key points

Anthropic's parallel publications: economy, misuse, and military-relevant capability

Anthropic released "Scenarios for our Economic Future," its contribution to the ongoing debate over how advanced AI could reshape economic trajectories — part of the company's continuing effort to frame policy discussion around AI's macroeconomic effects rather than leave the conversation to speculation.

On the same day, Anthropic put out its September 2026 threat-intelligence and misuse disclosure alongside a distinct assessment titled "Measuring AI capabilities in intelligence targeting and conventional weapons." Publishing a defense-relevant capability evaluation in the same window as its routine misuse reporting signals that Anthropic is treating dual-use, national-security-adjacent capability measurement as its own reporting category rather than folding it into general misuse disclosures.

Sources: Scenarios for our Economic FutureAnthropic · Threat IntelligenceAnthropic · Detecting and countering misuse of AI: September 2026Anthropic · Measuring AI capabilities in intelligence targeting and conventional weaponsAnthropic

Claude Code's rapid-fire releases: effort caps, gateway hardening, and a self-inflicted bug fixed

Four Claude Code point releases landed in quick succession. 2.1.267 introduced a maxEffortLevel setting that caps the effort level across every provider, including Bedrock, Vertex and Foundry, while still letting individual users choose a lower level. 2.1.268 focused on self-hosted gateways: pricing declared in gateway.yaml now flows through to signed-in clients so that /cost and telemetry match the actual spend meter, and administrators gained a gatewayInternalNetworks setting to scope /login to their organization's own public IPv4 block. 2.1.269 added claude plugin eval for scored, reproducible plugin testing, an /output-style command usable even in headless and cloud sessions, and an optional diff of files changed by Bash commands.

The same 2.1.268 release also added two security warnings for gateway operators: one at startup if access_control.allow_cidrs is left empty, and a one-time alert the first time a request arrives from a public address — a direct nudge toward closing off accidental public exposure of self-hosted gateways. And 2.1.270 quietly fixed a regression from 2.1.269 itself, where read-only git commands in Bash had started unexpectedly asking for permission again after a session had been running for a while.

Sources: Claude Code 2.1.267 release notesClaude Code · Claude Code 2.1.268 release notesClaude Code · Claude Code 2.1.269 release notesClaude Code · Claude Code 2.1.270 release notesClaude Code

Tighter human oversight for autonomous agents

Claude Managed Agents' permission policies now include an "auto" mode: the server itself evaluates each agent or MCP tool call and decides whether to run it, deny it, or pause for a human's approval, with the outcome recorded in an evaluation field on the corresponding tool-use events. The ant CLI gained a matching capability, ant beta:sessions connect, which attaches a terminal to a live Managed Agents session so an operator can follow it in real time, send messages, and approve or deny tool calls that are waiting on a decision — with a --web flag to serve the Console's session viewer locally instead.

Taken together, these are supervision primitives rather than new autonomy: the point of both features is to give a human a clear, real-time point of control over what an agent is about to do, which matters as more of these agents run unattended for longer stretches.

Sources: Claude Managed Agents permission policies now include autoAnthropic (release notes) · ant beta:sessions connectAnthropic (release notes)

Two outages, two different causes

Claude Cowork on Windows has been unable to run local commands since a Windows update released on September 8 left its workspace unable to reach the computer's drive — a cause Anthropic attributes explicitly to that Microsoft update rather than to a change on its own side. For most users chat and file reading and editing still work, but there is no in-app workaround, and neither restarting nor reinstalling helps; Microsoft has developed a fix and is working to release it, and Anthropic says it will post an update when that happens.

Separately, and unrelated to the Windows issue, Anthropic identified the cause of elevated error rates on requests to Claude Mythos 5.1 and Claude Fable 5.1 and resolved it the same day.

Sources: Degraded functionality for Claude Cowork on Windowsstatus.claude.com · Elevated errors for Claude Mythos 5.1 and Claude Fable 5.1status.claude.com

This edition is an original synthesis written by Claude from aggregated news — Anthropic's own sources first (release notes, status, newsroom, research, engineering), then the press, Hacker News, Reddit and GitHub, under the editorial supervision of Héra SASU. Every fact links to its article, publisher named. See the live feed →

Claude News is published by Héra SASU. Independent media, not affiliated with Anthropic.