Claude's memory now follows you into its Cowork agent
Anthropic just turned Claude's memory into one system shared between chat and its Cowork agent — with new privacy dials to match. Plus: Waymo picks Munich for its first EU robotaxi beachhead.
Anthropic has unified the memory behind Claude chat and Claude Cowork, so what the assistant learns about you in conversation now carries into the agentic desktop product, and back again. Users can see everything Claude remembers, topic by topic, and edit or delete any item; memories are written while the chat happens rather than summarized afterward, so mention a deadline moved to September and the next conversation already knows — no "remember this" required.
The privacy mechanics are the part worth reading twice. Sensitive subjects — health concerns, race, ethnicity, gender identity, religion, political beliefs — stay out of memory by default and are stored only if you flip a toggle. Even with the toggle on, some things are never saved: Social Security numbers, government IDs, criminal history, immigration status, and anything that would violate Anthropic's usage policy. As ZDNET's writeup frames it, chat history becomes available to Cowork unless users opt out — the convenience is the default, and staying private takes action.
That trade is the right one to argue about. Agents live on context: we covered Cowork's arrival in Chrome earlier this summer — Claude Cowork arrives in Chrome with skills and plugins — and an agent that already knows your manager's report preferences, your team's metric definitions, and your writing voice stops being a demo and starts carrying multi-week work. But merged memory also pools personal data where a work tool can reach it, which makes those sensitive-category defaults load-bearing rather than cosmetic. It's Anthropic's clearest answer yet to OpenAI's memory push: the assistant that never needs re-briefing wins the daily habit.
Waymo is heading to Germany — its first foothold inside the EU. The Alphabet unit said Tuesday it will begin phased testing in Munich in the coming weeks, building high-definition street maps, training autonomous specialists, and tuning its software to the city's driving quirks, with public rides targeted for the end of 2027. Munich becomes Waymo's third international market after London and Tokyo, still in testing themselves. Co-CEO Tekedra Mawakana called Munich "a world-class hub for mobility and engineering," and the company brings receipts: operations in 11 U.S. cities with 19 more planned, and nearly 221 million fully autonomous miles logged as of March. Europe is the real prize — heavily regulated, so far robotaxi-free, and no longer uncontested, since Pony.ai and Uber launched the continent's first driverless service earlier this month — Pony.ai and Uber launch Europe's first robotaxi service with 2,000 vehicles.
What to watch: whether EU privacy regulators read merged chat-agent memory generously — an always-on assistant that pools sensitive topics is exactly the profile GDPR scrutiny was built for.
Would you let your chat assistant's full memory flow into the agent working on your files, or keep those worlds separate? Tell us in the comments.
Sources: SiliconANGLE · ZDNET (via Techmeme) · CNBC · Waymo blog
