Claude's Memory Now Syncs Everywhere by Default — With SSNs and Criminal History Permanently Excluded

Anthropic has unified memory across Claude’s chat interface and Claude Cowork, syncing in real time as a user works across both — a saved preference or project detail from a chat now shows up inside Cowork sessions without the user re-explaining it, and vice versa. Every piece of saved information lives in an editable Topics list the user can review, edit, or delete directly, and memory is on by default across Free, Pro, and Max plans, on web, desktop, and mobile.

The default-exclusion list is the more consequential engineering decision here. By default, Claude excludes health information, religion, politics, and gender identity from what it remembers — categories a user can opt back into if they want continuity on those topics — while Social Security numbers, government IDs, criminal history, and immigration status are never stored, full stop, regardless of what a user sets. That’s a harder boundary than most consumer AI memory systems draw, and it’s specifically the boundary enterprise buyers ask about first when evaluating whether an AI tool can sit anywhere near HR, legal, or compliance workflows.

The design mirrors a pattern already visible in how Anthropic ships default-safety behavior elsewhere in the product line: Claude Code’s auto mode, made default in August, was tested against 1,053 paid users and found humans caught only 13.6% of dangerous commands manually, while auto mode’s automated blocking caught 89% of them — the same underlying bet that a well-designed default outperforms relying on the user to manually manage risk. Anthropic’s own usage data gives context for why memory continuity matters commercially: an analysis of roughly 1.2 million Cowork sessions across more than 600,000 organizations found business process and operations work was the largest single use category at 33.4% of sessions, more than triple software development’s 8.7% — exactly the kind of recurring, context-heavy work where memory continuity compounds in value session over session.

For enterprises evaluating AI vendors on data-governance maturity, the specific exclusion list — not just the existence of a privacy policy — is now a fair diligence question to ask every vendor, not only Anthropic.