Claude News
FR
Live

Claude, the essentials — edition of October 18, 2026

Anthropic gives METR employee-like access as independent evaluation widens

By ·

Anthropic says it has started giving a team from the nonprofit METR employee-like access, and METR will publish its own risk assessments. The same week brought Opus 5.5 fast mode out of research preview, a new Compliance API endpoint, and three Claude Code releases.

Key points

Independent evaluators get deeper access

In a post dated October 15, Anthropic said it has begun giving a team from METR, a nonprofit that evaluates frontier AI models, employee-like access. METR will publish the results of its risk assessments. Anthropic describes this as following a commitment made the previous month to stronger independent, third-party assessment. It also says it signed the White House Accord on Super Intelligence, in which companies pledged to partner with independent external auditors or evaluators. Anthropic adds that it has worked with METR over the past year on several reviews of its Risk Reports, and that SecureBio and Accenture are also acting as evaluators. It is in early discussions with several other organizations, including Hugging Face.

Anthropic's stated reasoning is that it wants to be publicly accountable but cannot publish everything relevant to its safety work. Some of that information could help malicious actors, and some is commercially sensitive, such as compute or training information. So it gives evaluators deep access and has them share evidence and reasoned conclusions. The stated aim is to close the gap between what evaluators understand about a frontier company's systems and what the company itself understands. It sets four requirements for evaluators: expertise in frontier-model assessment, including how evaluations can go wrong, as with alignment faking. They also need independence, meaning no equity in the company and no financial or operational dependence on it. Beyond that, they need enough access, including visibility into training and safeguards, and the freedom to publish their observations. Anthropic compares the model to bank supervision, where regulators place resident examiners inside the largest banks, while noting the analogy is not exact.

Sources: An update on independent evaluators — Anthropic

Opus 5.5 fast mode graduates; compliance tooling widens

Anthropic's API release notes for October 15 say fast mode for Claude Opus 5.5 has left research preview, so access no longer has to be requested. Requests no longer need the fast-mode-2026-02-01 beta header, and requests that still send it keep working. Fast mode for Opus 5.5 is also now available on Claude in Microsoft Foundry for deployments hosted on Azure, extending the feature to a second route to the model.

On the administration side, the Compliance API now returns the text each organization member saved as Instructions for Claude in claude.ai and as Global instructions in Cowork. The feature is in beta for Claude Enterprise organizations and is read through GET /v1/compliance/organizations/{org_uuid}/user_preferences. The notes frame it as an addition to the existing Compliance API access, giving administrators visibility into the standing instructions members have set.

Sources: Anthropic API release notes, October 15, 2026 — Anthropic (release notes)

Claude Code 2.1.300 to 2.1.302: configuration and a closing date

Version 2.1.302 adds the CLAUDE_CODE_AUTO_MODE_MODEL variable, which selects the model used by the client-side auto mode classifier. The release notes specify that it does not change the model for server-side auto mode. The same release adds an Idle compaction row to /config so users can turn idle compaction off or back on, and the notices shown after an idle compaction now explain how to disable it. It also adds a notice to /design-sync and /design-login stating that standalone Claude Design (claude.ai/design) closes by December 14, 2026.

Version 2.1.301 adds CLAUDE_CODE_RETRY_WATCHDOG_MAX_USAGE_LIMIT_WAIT_MS. With it, unattended retry mode (CLAUDE_CODE_RETRY_WATCHDOG) ends a request at a usage limit that resets too far away instead of waiting. The release also adds OpenTelemetry tracing guidance to the claude-api skill, including how to set up trace destinations with the ant CLI, and a % motion in vim mode that jumps to the matching bracket, alone or with d, c or y. Version 2.1.300 adds virtualText to a mod's prompt decorations, which is text drawn in the prompt box that never enters the draft or the submitted prompt.

Sources: Claude Code 2.1.302 — Claude Code · Claude Code 2.1.301 — Claude Code · Claude Code 2.1.300 — Claude Code

New Anthropic writing on science and work

Anthropic's feed lists five further pieces. Two concern technical subjects: "Designing more efficient quantum circuits with Claude" and "Claude reveals ECDSA curve origins". The other three concern work and expectations: "The AI expectations gap", "How employment is changing at Anthropic" and "Working at the frontier shapes expectations for the future".

Only the titles reached us for these items, with no summaries. We therefore make no claim about their findings and recommend reading them directly before drawing conclusions.

Sources: Designing more efficient quantum circuits with Claude — Anthropic · Claude reveals ECDSA curve origins — Anthropic · The AI expectations gap — Anthropic · How employment is changing at Anthropic — 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.