Meta's Muse builds a page for everyone in your life

A fresh round of uncomfortable evidence about what AI agents quietly keep on you — and the first concrete sign that AI-generated slop is now breaking security programs, not just feeds.

Meta's Muse builds a page for everyone in your life. Extracted system prompts for Meta's consumer agent show an instruction to compile "a page for every person in the user's life," with an hourly background job filling sections labeled Facts, History, The relationship, In common, Open threads, and Strengthening — "a reason to call, a date worth remembering, something they said to circle back on." WIRED's report lands on top of two independent analyses of the same leaked VM export: the Future Society review flags the high-severity finding that these pages cover "people who never signed up for Muse or agreed to it," and researcher Peter James's teardown of the 6.8 GB export documents the background jobs that maintain them. Meta spokesman Daniel Roberts said Muse gathers the material "based on public information and from what you've chosen to share" — which is precisely the problem: the people being profiled never chose anything. Millions have already downloaded Muse, and the privacy cost extends past the user to everyone the user knows. Muse's record with our readers is already long — we covered Meta's Muse assistant shipped with a local zero-day — but profiling non-consenting third parties is a new low-water mark for consumer agents.

Google freezes its open-source bug bounty over AI spam. As of October 1, Google is no longer accepting product-vulnerability submissions to its OSS Vulnerability Reward Program, citing "a significant rise in automated submissions, the vast majority of which are not valid," and won't revisit the format until an update planned for Q1 2027. Supply-chain reports and anything filed before the cutoff still count, but Tom's Hardware reports engineers are wading through thousands of sloppy filings — many of them outright hallucinated bugs — instead of fixing real flaws. This is the AI-slop war reaching its logical endpoint: the volunteer security work that keeps the open-source stack alive is being drowned by agents, and only a company with Google's resources can even attempt triage. Smaller maintainers have no such backstop.

Musk renames SpaceXAI to SpaceXSI. In an X reply Sunday, Elon Musk confirmed SpaceX's AI unit will become SpaceXSI — "No more AI, SI. It's better" — following Trump's push to swap "artificial" for "super." The Information, Gizmodo and Forbes all picked up the rebrand within hours. It's one executive's post, but it's also the clearest sign yet that the White House's "super intelligence" branding is propagating through the industry by social pressure rather than policy.

What to watch: whether Meta extends the same "public information" logic to a formal privacy policy — and how many other bounty programs pause before Google did.

Would you let an AI agent maintain dossiers on your friends and family? Tell us in the comments.