FR
Live

Claude, the essentials — edition of November 8, 2026

Access controls tighten as Anthropic spotlights Claude's role in science

By ·

Anthropic added two administrative security controls in beta, and Claude Code's auto mode now behaves differently when it can't reach server-side checks. Alongside these, Anthropic published a group of pieces on scientific research, agent memory and verifiable inference.

Key points

Administrative controls for organizations

Two additions in the API release notes dated November 5 are aimed at administrators. The Claude Console now offers an IP allowlist, in beta. Requests to covered services from addresses not on the list are refused. Anthropic frames the purpose as limiting where a leaked API key can be used. Administrators can add IPv4 and IPv6 addresses or CIDR ranges, with a cap of up to 1,000 per organization.

The second addition is for Claude Enterprise (claude.ai). The Admin API, also in beta, can now sign a member out of Claude everywhere. The member is not removed: they stay in the organization and can sign in again. This makes it a session-revocation tool, not an offboarding one. Together, the two controls address two different exposures, stolen keys and sessions that are still open.

Sources: Claude API release notes, November 5, 2026: IP allowlist in the Claude Console (beta) — Anthropic (release notes) · Claude API release notes, November 5, 2026: sign a member out everywhere with the Admin API (beta) — Anthropic (release notes)

Claude Code: a stricter auto mode and more background work

The most consequential change in Claude Code 2.1.319 concerns auto mode. When telemetry or feature flags are off, including behind LLM gateways, calls the server can't check are now asked about or denied, with no fallback to the built-in classifier. The release note names CLAUDE_CODE_AUTO_MODE_SERVER=0 as the setting for keeping the classifier in use. The truncated note also indicates that Bedrock, Vertex and Foundry are unchanged. A second entry covers gateway setups: CLAUDE_CODE_AUTO_MODE_SERVER=1 turns on the server-side check even when CLAUDE_CODE_DISABLE_EXPERIMENTAL_BETAS is set. If managed settings set that variable, they must also set the 1.

The same release adds support for MCP tasks in the terminal. When an MCP server runs a tool call as a task, Claude Code moves it to the background right away and tells Claude the result when it finishes. Version 2.1.318 adds an assume_role option on mantle upstreams of the Claude apps gateway, for a role in another AWS account and for per-developer AWS cost attribution. It also records your answer when Claude Code asks to turn on the mods it writes in a session, and it adds an effort option to a mod's $.model.fork so a quick side job can run at low effort whatever effort the session uses. Version 2.1.317 adds a warning to /schedule when a new cloud scheduled task gets connectors you did not ask for, with instructions for removing them.

Sources: Claude Code 2.1.319: auto mode with telemetry or feature flags off — Claude Code (GitHub) · Claude Code 2.1.319: MCP tasks in the terminal — Claude Code (GitHub) · Claude Code 2.1.318: assume_role, mods and effort option — Claude Code (GitHub) · Claude Code 2.1.317: /schedule connector warning — Claude Code (GitHub)

Science takes the foreground

Anthropic published a cluster of pieces on research use. One looks at how researchers use Claude for scientific discovery, and another is titled Teaching AI to build scientific models. A third, Investigating vault particles with Claude, is paired in Anthropic's feed with a 2026 paper by Athukoralage et al. on prokaryotic homologs of the major vault protein forming engineerable vault nanocapsules. The items we received carry titles but no body text. We can therefore say what the subjects are, but not what results or methods the pieces report or what part Claude played in the vault work.

Taken together, they suggest Anthropic is presenting Claude as a tool inside research workflows, from building models to working on a specific biological question. A precise reading of that role will have to wait for the full texts.

Sources: AI for scientific discovery: How researchers use Claude — Anthropic · Teaching AI to build scientific models — Anthropic · Investigating vault particles with Claude — Anthropic · Athukoralage et al. 2026, Prokaryotic homologs of the major vault protein form engineerable vault nanocapsules — Anthropic

Agents: memory, games and verifiable inference

Three other Anthropic pieces concern how agents and models are built and checked. The components of agent memory addresses a design question for long-running agents. Project EVE: Claude plays MMORPGs describes an experiment in which Claude plays massively multiplayer online games. Progress on provable inference for large language models covers work toward making a model's outputs verifiable. As with the science items, only the titles reached us, so we make no claims about their findings.

Read as a group, the three subjects are persistence over time, open-ended environments and verifiability. Each bears on how far agents can be trusted with extended tasks.

Sources: The components of agent memory — Anthropic · Project EVE: Claude plays MMORPGs — Anthropic · Progress on provable inference for large language models — Anthropic

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.