Anthropic opens a Singapore office and signs an AI-safety MOU with the government

Singapore becomes a local base and a policy partner

Anthropic is expanding its Asia-Pacific operations with a new Singapore office, the fifth in the region, and a Memorandum of Understanding with Singapore's Ministry of Digital Development and Information. The company describes Singapore as one of its most active markets anywhere. It says Claude already powers business and government services used by millions of people every day, and it names GovTech Singapore, Grab and DBS. Dale Finlay, General Manager of ASEAN, said customers there hold AI to an extremely high standard on reliability and security, and that a team on the ground lets Anthropic get closer to organizations putting Claude to work on their hardest problems.

The MOU was signed by Kai Fong Chng, the ministry's Permanent Secretary, and Anthropic co-founder Ben Mann. The signing took place on the sidelines of Singapore International Cyber Week, and Minister Josephine Teo and Michael Sellitto, Anthropic's Head of APAC Policy, witnessed it. It 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. Mann cited Singapore's long record of investment in model testing and evaluation, and said Anthropic wants to help the transition to powerful AI go well in Singapore, across Asia-Pacific and worldwide. Chng framed the partnership as part of the government's effort to scale up responsible AI adoption in companies and agencies.

Sources: Singapore office and government MOU — Anthropic

Safety, security and engineering publications from Anthropic

Several Anthropic publications appear in the day's feed, and for these the material available to us is limited to titles. One is headed "AI models may be capable of self-replicating without access to a datacenter". Another is "Investigating unintended model actions in our evaluations and internal use". A third is "Strengthening our defenses against distillation attacks". The Frontier Red Team is listed with an item titled "OSS Scanner". We do not have the contents of these pieces, so we report only their subjects and make no claim about their findings or conclusions.

One further item, "Redesigning source control at Anthropic", points to an internal engineering topic, and again only its title is available. Read alongside the Singapore MOU, which names AI safety, security and model evaluation as areas of cooperation, the titles show Anthropic publishing on model behavior, defenses and security tooling in the same period.

Sources: AI models may be capable of self-replicating without access to a datacenter — Anthropic · Investigating unintended model actions in our evaluations and internal use — Anthropic · Strengthening our defenses against distillation attacks — Anthropic · OSS Scanner — Anthropic (Frontier Red Team)

Claude Code and Managed Agents: more control over context, hooks and workflows

Claude Code 2.1.299 adds idle compaction in the terminal. A conversation of 200K or more tokens that sits idle for about 54 minutes is compacted before its one-hour prompt cache expires, so the next message does not reprocess everything. The threshold can be changed with CLAUDE_CODE_IDLE_COMPACT_MIN_TOKENS, and "idleCompaction": false in settings.json turns the feature off. Both settings require v2.1.290 or later. The same release 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 2.1.296 focuses on managed and multi-agent setups. It adds a code key to the Claude apps gateway's managed.policies[] that applies the same settings as cli and also applies in Claude Desktop's Code tab, where it turns on the desktop's gateway mode. It adds autoCompactWindow to subagent frontmatter and --agents definitions so a subagent can auto-compact earlier than the main conversation. It also adds CLAUDE_CODE_WORKFLOW_SUBAGENT_MODEL to run every workflow agent on one model while other subagents keep theirs. Version 2.1.295 adds onFailure: "block" for command and HTTP hooks. A hook that cannot start, times out or exits with an unexpected code then blocks the action instead of letting it through. It also supports the Program Status Protocol (OSC 7501), so compatible terminals can show whether Claude Code is working, waiting or done.

On the API side, the release notes dated October 9 say a Claude Managed Agents agent can now use dynamic workflows, in beta behind the managed-agents-2026-04-01 header. For work with many pieces, such as reviewing hundreds of documents, the agent can write a workflow, which the notes describe as a program that runs many agents in phases. The workflow concept now appears in both Claude Code settings and Managed Agents.

Sources: Claude Code 2.1.299 — Claude Code · Claude Code 2.1.296 — Claude Code · Claude Code 2.1.295 — Claude Code · Claude Managed Agents dynamic workflows (API release notes, October 9, 2026) — Anthropic (release notes)

Status: usage data errors on platform.claude.com, now resolved

The status page lists an incident titled "Elevated errors on platform.claude.com" and marks it resolved. It concerned errors affecting usage data on the Usage and Rate limits pages of platform.claude.com and in the Admin API usage report. According to the entry, a mitigation for the Usage page and the Admin API usage report was rolling out. Once it was in place, the most recent usage could take about 30 minutes to appear there. Anthropic states that core functionality was not affected.

Sources: Elevated errors on platform.claude.com — status.claude.com