Claude News
FR
Live

Claude, the essentials — edition of November 4, 2026

Claude Code adds guardrails and gateway controls; the API adds message threads

By ·

Three recent Claude Code releases give administrators more control over the Claude apps gateway and add several checks before the tool acts on a user's behalf. On the API side, a beta lets the server hold conversation state, and the release notes describe text watermarking for three models.

Key points

Claude Code gives administrators finer control over the apps gateway

Three recent Claude Code releases (2.1.314, 2.1.316 and 2.1.318) each add a control to the Claude apps gateway, which sends developers to configured upstreams. Version 2.1.314 lets an upstream carry an optional list of IdP groups, so that only developers in those groups are sent to it, including on failover. Version 2.1.316 adds pooled group budgets: one monthly budget shared by a group's members, applied on top of each member's own limit and set with pooled_budget on the spend limits admin API. Version 2.1.318 adds assume_role on mantle upstreams, for a role in another AWS account and for per-developer AWS cost attribution.

Taken together, the additions cover who is routed to which upstream, how much a group may spend, and how AWS usage is attributed to individual developers. These are concerns of administrators managing a group of developers rather than a single user. The same releases also carry smaller options. Version 2.1.316 adds CLAUDE_CODE_DISABLE_ULTRATHINK, which turns off the ultrathink keyword, and CLAUDE_CODE_DISABLE_LOOPED_CROSS_CERT_DROP, which keeps system cross-certificates that certify each other in a loop, for servers whose private certificate authority needs one. Version 2.1.318 records the user's answer when Claude Code asks to turn on the mods it writes in a session. It also 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; the result reports the effort sent.

Sources: Claude Code 2.1.314 — optional groups list on Claude apps gateway upstreams — Claude Code · Claude Code 2.1.316 — pooled group budgets, CLAUDE_CODE_DISABLE_ULTRATHINK and cross-certificate option — Claude Code · Claude Code 2.1.318 — assume_role on mantle upstreams, mod answer record, effort option for $.model.fork — Claude Code

More checks before Claude Code acts on a user's behalf

Other entries in the same releases concern what Claude Code does for the user. Version 2.1.314 adds permissions.blockWritesOutsideWorkingDirectories, under which Write, Edit and NotebookEdit refuse paths outside the working directories in every permission mode. Version 2.1.317 adds a check that asks before a web or shell tool call sends the account email address to the network. It also adds onFailure: “block” for Agent SDK hook callbacks, so that a callback that fails blocks the action instead of letting it run.

Version 2.1.317 also adds a warning in /schedule when a new cloud scheduled task gets connectors the user did not ask for, with instructions for removing them. The entry does not say how such connectors come to be attached. What links these changes is that each gives the user or the integrator a decision point or a way to fail closed: a path boundary, a confirmation before an identifier leaves the machine, a failure mode that stops rather than proceeds, and a notice about connectors nobody requested.

Sources: Claude Code 2.1.314 — permissions.blockWritesOutsideWorkingDirectories — Claude Code · Claude Code 2.1.317 — email address check, onFailure “block” for hook callbacks, /schedule connectors warning — Claude Code

The Messages API moves conversation state to the server, and the notes document watermarking

The API release notes for November 2 describe message threads, in beta behind the message-threads-2026-08-12 beta header. On the Messages API, long conversations and agent loops can send smaller requests and get a lower time to first token, because the API keeps the conversation on the server, which is what allows later requests to be smaller.

The notes dated November 3 concern provenance. They state that text generated by Claude Mythos 5, Claude Opus 4.7 and Claude Sonnet 4.6 carries Anthropic’s text watermark on the platforms listed in “How Claude marks AI-generated content”. The entry also mentions a further condition for Claude Sonnet 4.6 requests that set temperature, top_p or top_k. The excerpt available to us breaks off before stating it, so we do not describe it.

Sources: Anthropic API release notes — November 2, 2026 (message threads) — Anthropic (release notes) · Anthropic API release notes — November 3, 2026 (text watermark) — Anthropic (release notes)

Research posts on provable inference and misalignment, and a Residency call

Two research items come from Anthropic's own channels. One is titled “Progress on provable inference for large language models”. The other, labelled Alignment Science blog, is titled “Automated Subtle Emergent Misalignment Research”. The material supplied for both is limited to the titles, so we do not summarize findings or claims the posts may make.

A third item, also labelled Alignment Science blog, announces that applications are open for the Anthropic Residency, aimed at established technical leaders. No further terms are available to us, such as duration, location or selection criteria, and none should be inferred.

Sources: Progress on provable inference for large language models — Anthropic · Automated Subtle Emergent Misalignment Research — Anthropic Alignment Science blog · Anthropic Residency: applications open for established technical leaders — 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.