Claude can now show users not only what they asked it to do, but also the work patterns those conversations reveal. Anthropic launched Reflect on July 9, 2026, with recaps covering recurring subjects, conversation activity, peak usage times, and tasks users hand off to the assistant.
Reflect is limited to Free, Pro, and Max accounts on Claude’s web and desktop apps. Team and Enterprise accounts are excluded, leaving IT teams to consider how employees may create persistent summaries of company work through personal Claude accounts outside managed controls.
Claude memory supplies the context behind Reflect
Users open the feature through Settings > Reflect. Anthropic’s monthly recap documentation says it initially displays the previous month, with options for the current month, past three months, past six months, or past year.
The dashboard shows conversation totals, peak activity times, common topics, and observations tied to Anthropic’s four AI fluency areas: delegation, description, discernment, and diligence.
Reflect relies on the chat history used by Claude memory. Anthropic’s memory documentation says non-project conversations are summarized into a general memory synthesis that updates every 24 hours. Each Claude project has a separate memory space and summary.
Pausing memory hides the recap but preserves existing memory. Resetting memory permanently deletes general and project memories.
Reflect excludes incognito chats, conversations using health integrations, and activity in Claude Code and Claude Cowork. The Cowork exclusion comes as Anthropic expands Claude Cowork across web and mobile, allowing longer-running tasks involving files, email, calendars, and connected services.
Reflect does not incorporate raw emails or files retrieved from connected services, although text Claude produces about that material inside a chat may be included.
Sensitive or distress-related subjects are not supposed to lead the recap or appear with itemized percentages. Anthropic also warns that generated summaries, statistics, and topic proportions may occasionally be inaccurate.
Break reminders and quiet hours appear beside Reflect but operate independently. The optional Time and focus controls do not require users to open the recap, and neither setting blocks continued Claude use.
Enterprise controls do not extend to personal accounts
Organizations cannot deploy Reflect through a managed Team or Enterprise tenant. The near-term concern is whether employees enable memory on personal Claude accounts and create persistent summaries of company information outside approved access, retention, and offboarding controls.
Enterprise owners can disable memory organization-wide. The capability is enabled by default, after which users control their individual settings. Disabling it at the organization level permanently deletes existing memory synthesis data for all users. Team plans lack an equivalent organization-wide control.
Memory follows existing enterprise chat-retention policies and is included in standard data exports. Deleted conversations are removed from the synthesis, although updates may take up to 24 hours. Organization-level setting changes are logged; individual memory edits are not.
Company policies should prohibit credentials, regulated personal data, privileged legal material, incident details, unreleased product information, and sensitive source code from personal AI accounts. Reports of undocumented checks inside Claude Code also reinforce the need to understand what AI tools inspect, modify, and retain.
Reflect remains a consumer-account beta, but its memory requirement gives enterprises another reason to distinguish approved workplace accounts from personal AI use. Persistent summaries should be covered by the same policies that govern prompts, uploaded files, retention, and employee offboarding as policymakers and researchers push for stronger AI oversight.
Read more: The first reported agentic ransomware attack shows why identity controls, monitoring, and human oversight must advance alongside increasingly autonomous AI systems.