Claude, the essentials — edition of October 16, 2026
Anthropic gives METR employee-like access as outside scrutiny of its models grows
Anthropic has begun giving a team from the nonprofit evaluator METR employee-like access, and METR will publish its own risk assessments. Around that announcement, the company also set out its midterm election safeguards, opened a Singapore office, and shipped more observability and compliance tooling.
- Anthropic has started giving a METR team employee-like access. METR will publish the results of its risk assessments.
- SecureBio and Accenture are also working as evaluators. Anthropic is in early discussions with several others, including Hugging Face.
- In a sample of about 250,000 US Claude.ai conversations, about 0.51% were specifically about elections or voting, and about 0.01% asked whom or what to vote for.
- Anthropic opened its fifth Asia-Pacific office, in Singapore, and signed an MOU with the country's digital ministry that covers AI safety, security and model evaluation.
- Fast mode for Claude Opus 5.5 is out of research preview. OpenTelemetry export is in beta for the API, and the Compliance API now exposes members' saved instructions.
Independent evaluators move inside the perimeter
Anthropic announced on October 15 that it has started giving a team from METR employee-like access. METR is a nonprofit that evaluates frontier AI models, and it will publish the results of its risk assessments. Anthropic ties the step to a commitment it made last month to stronger third-party assessment, and to its signature on the White House Accord on Super Intelligence, in which companies pledged to partner with independent external auditors or evaluators. It says METR has already worked with it on several reviews of its Risk Reports over the past year. SecureBio and Accenture are also acting as evaluators, and discussions are at an early stage with several other organizations, including Hugging Face.
Anthropic's stated reason is a limit on transparency. It says it cannot publicly share all the information relevant to its safety work, because some of it could help malicious actors and some is commercially sensitive, such as compute or training information. Its alternative is to give evaluators deep access and have them publish evidence and reasoned conclusions. It lists four requirements: expertise in assessing frontier models, independence (no equity in the company and no financial or operational dependence on it), enough access to see how models are trained and how safeguards work, and the freedom to publish. It compares the arrangement to bank supervision, where resident examiners work inside the largest banks, and adds that the analogy is not exact.
Sources: An update on independent evaluators — Anthropic
Elections and a new regional partner
For the US midterms, Anthropic published an analysis of about 250,000 US Claude.ai consumer conversations from September 21 to 27, 2026. It classified them with a privacy-preserving method. Conversations touching politics, public policy or social issues made up about 5.75% of US conversations on Claude.ai, and about 0.51% were specifically about elections or voting. Requests for a recommendation on whom or what to vote for were rare, at about 0.01%. Anthropic notes that most requests are informational, such as asking where candidates stand on an issue. Requests that violated its election integrity policies appeared in about 0.002% of conversations. The company expects the share of political and election-related queries to rise as Election Day approaches.
On October 14, Anthropic opened an office in Singapore, its fifth in Asia-Pacific. It also signed a Memorandum of Understanding with the Ministry of Digital Development and Information, on the sidelines of Singapore International Cyber Week. The company calls Singapore one of its most active markets anywhere and names GovTech Singapore, Grab and DBS as organizations using Claude. The MOU sets out cooperation on AI safety, security and model evaluation, research on AI adoption and its impact, and responsible adoption across the public and private sectors. Co-founder Ben Mann signed it for Anthropic, and the ministry's Permanent Secretary, Kai Fong Chng, signed for the government.
Sources: Preparing for the 2026 US midterms — Anthropic · Anthropic opens Singapore office and signs MOU with the Government of Singapore — Anthropic
API: fast mode graduates, tracing and compliance data arrive
According to the API release notes, fast mode for Claude Opus 5.5 is out of research preview and no longer requires an access request. 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.
Two other changes concern visibility into usage. On October 14, Anthropic launched OpenTelemetry export in beta, so teams can trace Messages API requests and Claude Managed Agents sessions in their own observability backend. The Claude SDK emits spans in Python, TypeScript, C#, Go, Java, PHP and Ruby. On October 15, the Compliance API began returning the text each organization member saved as Instructions for Claude in claude.ai and as Global instructions in Cowork. This is in beta for Claude Enterprise organizations and is served from the user_preferences endpoint.
Sources: Claude API release notes, October 15, 2026 — Anthropic (release notes) · Claude API release notes, October 14, 2026 — Anthropic (release notes)
Claude Code 2.1.299 to 2.1.301: cache economics and telemetry
Version 2.1.299 adds idle compaction in the terminal. A conversation of 200K or more tokens that has been idle for about 54 minutes is compacted before its one-hour prompt cache expires, so the next message does not reprocess the whole conversation. The minimum can be changed with CLAUDE_CODE_IDLE_COMPACT_MIN_TOKENS, and setting idleCompaction to false in settings.json turns the feature off. Both need v2.1.290 or later. The release also adds CLAUDE_CODE_DISABLE_FILE_MENTIONS, which makes Claude receive the paths of @ file mentions to read with its file tools instead of the files' contents.
Version 2.1.300 adds cache_creation_5m_tokens and cache_creation_1h_tokens to the OpenTelemetry api_request event, which splits cache writes by lifetime. For mods, it adds virtualText to prompt decorations, text drawn in the prompt box that never enters the draft or the submitted prompt, and $.ui.view({ agentId }), which shows an agent's transcript in the main view on request. Version 2.1.301 adds CLAUDE_CODE_RETRY_WATCHDOG_MAX_USAGE_LIMIT_WAIT_MS, so unattended retry mode ends a request at a usage limit that resets too far away instead of waiting. It also adds OpenTelemetry tracing guidance to the claude-api skill, including setting up trace destinations with the ant CLI, and a % motion in vim mode that jumps to the matching bracket.
Sources: Claude Code 2.1.299 — Claude Code · Claude Code 2.1.300 — Claude Code · Claude Code 2.1.301 — Claude Code
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.