Anthropic spread its October news across four separate announcements rather than one event, and together they sketch a company pulling in two directions at once: shipping more autonomous agent behavior to paying users, while tightening the written rules about how its systems may be used.

Claude Duet: an agent that acts first (October 21)

The headline launch is Claude Duet, a new app in which Claude operates as a proactive agent — initiating work and following through on objectives rather than only responding inside a chat. It launched October 21 as a research preview for Pro and Max plan subscribers, which is both a capability statement and a scope statement: Anthropic is explicitly collecting feedback from its most engaged (and most expensive) users before any wider release.

The research-preview framing matters for expectations. Features can change weekly, some may not survive to general availability, and the usual advice applies — connect the minimum permissions you need, review what the agent does, and don't give it anything you wouldn't hand a new employee on day one.

Claude Haiku 5.5: the speed tier (October 7)

Two weeks earlier, on October 7, Anthropic released Claude Haiku 5.5, positioned as the fastest and most affordable member of the Claude 5 family. Haiku models are the workhorses — high-volume classification, extraction, support deflection, and the latency-sensitive paths of agent pipelines. A cheaper, faster Haiku generally shows up in customer bills before anyone notices a model change, which is the point.

A usage policy rewrite, effective November 12

On October 8, Anthropic published its 2026 Usage Policy update, which takes effect on November 12. The company paired the release with an unusual step: a public consultation on Claude's values, inviting outside input on the ethics questions that shape the policy. If you build on the API or operate Claude in a product, the November 12 date is the one to watch — usage policies are enforceable documents, and updates have historically arrived with enforcement tooling.

Frontier Academy: the $100M skills bet (October 2)

The month opened with a workforce announcement: Frontier Academy, a $100 million commitment to train 10,000 “Frontier Deployed Engineers” by the end of 2027. It's a supply-side move — the agent deployments Anthropic is selling need people who can run them, and the current market doesn't have enough.

Frequently asked questions

Can I try Claude Duet on the free plan?

No. Duet launched as a research preview for Pro and Max subscribers only.

What changes when the usage policy takes effect on November 12?

The updated policy governs how Claude may be used across consumer and API surfaces from that date. Developers with automated pipelines should re-read it before November 12 and check their use cases against the updated restrictions.

Is Haiku 5.5 available on the API?

Yes — Haiku models ship across Claude surfaces including the API; it is positioned as the speed and cost tier of the Claude 5 family.

Sources