Every item below was published by Anthropic on anthropic.com, X or LinkedIn in November 2026, listed newest first with its date and a link to the original post. The launches that change something for a European business running Claude also carry a short note from us.
The first week of November was about the practical side of agents: who can see what they do, who connects them to the apps, and how you measure whether they pay off.
Anthropic showed how its own team uses Claude Duet, the app where Claude keeps your day moving and sends a message when something needs your attention. It is still a research preview, so try it on one small team and read through what it did.
Sources: Claude on LinkedIn
Claude Enterprise gets a security center that shows Claude's settings for chat, Cowork and Claude Code on one page, compared with Anthropic's recommended configuration where one exists. Auto-connect is also in beta for Team and Enterprise: admins connect the organisation to an app once, and each member is connected the first time they use it.
Sources: Claude on LinkedIn, Claude on LinkedIn
Claude Cowork runs commands and code locally in an isolated virtual machine on your computer, and administrators can now run the organisation's Endpoint Detection and Response (EDR) there too. If your security policy requires every endpoint to show up in one panel, ask your IT team whether it covers Cowork.
Sources: Claude on LinkedIn
Anthropic described how agents can organise their memory between sessions, how AI can help build and compare scientific models, and how Claude designed three electronics prototypes that worked without circuit changes. Comcast and Booz Allen Hamilton also reached the same conclusion running Claude Mythos on their own code: finding vulnerabilities is the easy part and validating them is the bottleneck.
Sources: anthropic.com/research, anthropic.com/research, anthropic.com/research, Claude on LinkedIn
Claude designed three electronics prototypes: a speakerphone, a voice-controlled hub and a handheld pet. When Diode built them, all three worked without circuit changes.
AI can help researchers build, adapt and interpret scientific models. The key is to review many plausible models rather than a single best one.
Anthropic describes how AI agents can organise their memory between sessions through dreaming, which fixes mistakes, creates notes and builds knowledge wikis.
Finding vulnerabilities is now the easy part, validating them is the bottleneck. Comcast scanned 170 million lines of code and found a critical authentication flaw.
Connect the accounts you choose and Claude answers money questions in plain language, based on your own balances and transactions. Access is read-only.
A security center shows Claude's security settings for chat, Cowork and Claude Code on one page, compared with Anthropic's recommended configuration where one exists.
IT security teams get one page to check Claude's settings against a recommended setup. If you run Claude Enterprise, open the security center and list every setting that differs from the recommendation, then decide which differences are deliberate.
He designs every piece himself, and each one is unique. He also handles the bookkeeping, quotes, scheduling and website.
Start with how many reps use Claude each week, then what Claude drafts and how much the reps keep. After that, look at pipeline.
Claude helped Anthropic map the first vault particle in prokaryotes. Its function in the cells is still unknown.
Admins connect the organisation to an app once. After that, each member with a matching email address is connected in the app the first time they use it.
It keeps your day moving and sends a message when something needs your attention. This is how the Anthropic team uses it, both at work and outside it.
Duet is a research preview, so try it on one small team and read through what it did. Pick one recurring task, let it run for a week and note every time it asked you first.
The programme is time-limited, with placements of between six weeks and six months. Anthropic is looking for people who have led ambitious work in a range of technical fields.
Administrators can now run the organisation's Endpoint Detection and Response (EDR) here too.
Security policies that require every endpoint to show up in one panel can now cover Cowork as well. Ask your IT team whether your EDR policy already includes the Cowork virtual machine.
The article presents FlashProve, an open system from Anthropic that uses cryptographic proofs so users can verify that an answer really comes from a specific AI model.
Analytics cookies show which pages are read. Marketing stays off unless you tick it under Details.