A security post dated October 23rd describes approval checks, sensitive-message screening and isolated credentials for the research-preview agent. Anthropic says a follow-on penetration test of the cloud browser and sign-in form is underway.
By [RuntimeWire Staff](https://runtimewire.com/author/runtimewire-staff)
· Published
Primary source: [Claude by Anthropic](https://claude.com/resources/articles/claude-duet-safety-privacy-and-security)
Why it matters #
Claude Duet can read connected accounts and web pages, then prepare actions on a user's behalf. Anthropic's account describes where it places the controls: approvals and sensitive-message screening govern what leaves the system, while isolated machines keep passwords, tokens and browser cookies away from the model. The company also acknowledges that page content still reaches the agent and that its checks can miss things.
Anthropic published a security account of Claude Duet dated October 23rd, detailing approval checks, sensitive-message screening and how the agent's credentials are separated from its working machines.
When you purchase through links in our articles, we may earn a small commission. This doesn’t affect our editorial independence.