Anthropic AI Updates: November 4, 2026

1. Claude Code 2.1.318 Lets Cloud Sessions Start Without a Repository and Reports When a claude.ai Sign-In Expires

Anthropic released Claude Code v2.1.318 on the evening of November 3 (US time). Cloud sessions on Anthropic-hosted environments gain a No repository option, so a session can start without picking a repository, and now expose the built-in Chromium under the usual Chrome names (google-chrome and CHROME_PATH) so browser tools other than Playwright can find it. For unattended use, claude auth status now reports when a stored claude.ai sign-in ends (loginExpiresAt), claude -p JSON results carry a matching credential_expires_at field, and claude -p text output prints a one-line stderr warning starting 10 days before the sign-in expires. The Claude apps gateway accepts assume_role on mantle (Amazon Bedrock) upstreams, so it can use a role in another AWS account and attribute AWS costs per developer, and it now logs a warning when a Google access token from an external_account federation credentials file is about to expire. Mods gain an effort option on $.model.fork, letting a quick side job run at low effort whatever the session uses, and Claude Code now records the user’s answer when it asks to turn on mods it wrote during a session. Self-hosted runner orchestrators also receive a CLAUDE_RUNNER_SPAWN_REASON variable (for example runner_lost or resumed) in the spawn-runner hook environment, explaining why Anthropic requested a new runner. Source

2. Anthropic’s Text Watermark Is Now Live on Claude Mythos 5, Opus 4.7 and Sonnet 4.6

Anthropic added a November 3 entry to its Claude Platform release notes stating that text generated by Claude Mythos 5, Claude Opus 4.7 and Claude Sonnet 4.6 now carries its text watermark on the platforms listed in its content labeling article, putting into effect the extension to older models it announced to API customers by email in late October. The entry names three models; Claude Opus 4.6, the fourth model named in that email, is not listed. The new text watermark documentation says the mark adds no tokens or hidden characters, carries no information about the user or organization and needs no change to requests or response handling, and it keeps one caveat: on Sonnet 4.6, a request that sets temperature, top_p or top_k to a non-default value might return unmarked text, so teams that rely on the watermark should omit all three parameters, including any that a framework or client library sets for them. Source

3. Claude Code Product Lead Cat Wu Leaves Anthropic Weeks Before Its Planned IPO

Anthropic confirmed that Cat Wu, head of product for Claude Code and one of its two most visible leaders alongside creator Boris Cherny, left the company on Friday, October 30, Business Insider reports; Wu announced the departure on X on November 3 without giving a reason. Wu joined Anthropic in summer 2024, tested an early Claude Code prototype, appeared in the February 2025 research preview demo, and also oversaw Cowork, Anthropic’s agent for non-coding work. Business Insider notes that Thariq Shihipar, who also worked on Claude Code, announced his own departure last month, and that the exit comes as Anthropic heads toward an IPO expected later in November. Source