Claude, the essentials — edition of October 17, 2026
Anthropic gives METR employee-like access as it widens independent safety evaluation
Anthropic says it has begun giving a team from the nonprofit METR employee-like access and that METR will publish its risk assessments. The same week brought a measurement of how Americans use Claude for politics ahead of the midterms, plus a run of Claude Code and API changes.
- Anthropic has started giving a team from METR employee-like access, and METR will publish the results of its risk assessments.
- SecureBio and Accenture are also evaluators, and early talks are under way with other organizations, including Hugging Face.
- In a sample of about 250,000 US Claude.ai conversations, politics and policy made up about 5.75%, elections about 0.51%, and requests for a voting recommendation about 0.01%.
- Fast mode for Claude Opus 5.5 is out of research preview, and the Compliance API can now return members' saved instructions for Claude Enterprise organizations.
- Claude Code 2.1.301 and 2.1.302 add new controls, and 2.1.302 warns that standalone Claude Design closes by December 14, 2026.
Independent evaluators get deeper access
On October 15, Anthropic said it has started giving a team from METR, a nonprofit that evaluates frontier AI models, employee-like access, and that METR will publish the results of its risk assessments. The step follows a commitment to stronger third-party assessment made a month earlier. Anthropic also says it signed the White House Accord on Super Intelligence, in which companies pledged to partner with independent external auditors or evaluators. METR is not new to the company: Anthropic says they have worked together over the past year on several reviews of its Risk Reports. SecureBio and Accenture are also acting as evaluators, and early discussions are under way with several other organizations, including Hugging Face.
Anthropic's stated reason is that it cannot publish everything relevant to its safety work. Some of it could be useful to malicious actors, and some is commercially sensitive, such as compute or training information. Its answer is to give evaluators deep access and have them publish 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 names four requirements: expertise, independence (no equity in the company and no financial or operational dependence on it), enough access to see how models were trained and how safeguards work, and freedom to publish. It compares the arrangement to bank supervision, where resident examiners work inside the largest banks, while noting the analogy is not exact. The text available to us is cut off partway through that comparison, so we report nothing beyond it.
Sources: An update on independent evaluators — Anthropic
What Americans actually ask Claude about elections
Ahead of the US midterms, Anthropic used its Anthropic Insights tool to analyze about 250,000 US Claude.ai consumer conversations from September 21 to 27, 2026. Claude classified them using a privacy-preserving method. Conversations touching broadly on politics, public policy or social issues came to about 5.75% of the total, and about 0.51% were specifically about elections or voting. Anthropic expects both shares to rise as Election Day approaches. It describes most requests as informational, such as following the news, researching policy or candidates, and navigating government services.
Direct requests for a voting recommendation were rare: about 0.01% of US conversations. Because the group is so small, Anthropic reports the upper end of a 95% confidence interval for it. Requests that violated its election integrity policies appeared in about 0.002%. Anthropic says people more often ask questions that help them reach their own conclusions, such as where candidates stand on an issue. The post builds on an earlier update on its election safeguards, which covered measuring and preventing political bias, detecting misuse, and directing people to accurate voting information. The excerpt available to us covers the usage analysis, not the newer safeguard measures.
Sources: Preparing for the 2026 US midterms — Anthropic
More levers for operators and administrators
The two Claude Code releases of the week mostly add controls. Version 2.1.302 adds CLAUDE_CODE_AUTO_MODE_MODEL, which sets the model used by the client-side auto mode classifier. Anthropic specifies that it does not change the model for server-side auto mode. The release also adds an Idle compaction row to /config to turn idle compaction off or on, and the notices shown after an idle compaction now explain how to disable it. Its /design-sync and /design-login commands now carry a notice that standalone Claude Design (claude.ai/design) closes by December 14, 2026. Version 2.1.301 adds a variable so that unattended retry mode ends a request when a usage limit resets too far away, rather than waiting. It also adds OpenTelemetry tracing guidance to the claude-api skill, including setting up trace destinations with the ant CLI, and a % bracket-matching motion in vim mode.
On the API side, the October 15 release notes say fast mode for Claude Opus 5.5 is out of research preview. Access no longer needs to be requested, and the fast-mode-2026-02-01 beta header is no longer required, though requests that still send it keep working. Fast mode is also available on Claude in Microsoft Foundry for deployments hosted on Azure. For Claude Enterprise organizations, the Compliance API now returns, in beta, the text each member saved as Instructions for Claude in claude.ai and as Global instructions in Cowork, through a GET endpoint on the user_preferences path.
Sources: Claude Code 2.1.302 release notes — Anthropic (Claude Code on GitHub) · Claude Code 2.1.301 release notes — Anthropic (Claude Code on GitHub) · Claude API release notes, October 15, 2026 — Anthropic (release notes)
Other Anthropic publications in the day's feed
Four more Anthropic items appeared in the feed with headlines only, so we describe none of their content. Two are titled "Designing more efficient quantum circuits with Claude" and "Claude reveals ECDSA curve origins". Judging by the titles alone, both concern technical problems approached with Claude. The other two are titled "The AI expectations gap" and "How employment is changing at Anthropic".
We list them for completeness, and readers should consult the originals for what they actually say.
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.