WIRED found instructions for Muse to track people in a user’s life. Meta says users can erase that memory. The report does not test whether the erasure works.
Amazon says it no longer asks county officials to sign secrecy deals for data centers. WIRED asks whether the change covers contractors too. The pledge alone does not make every deal public.
Trillium Labs plans to publish AI experiments so others can check them, WIRED reports. Its founders favor openness. The article also raises the risk that some findings could be misused.
A Stratechery roundup argues that AI assistants could bring many online services together. That is an analyst’s view, not a count of users or proof that Muse leads the market.
Anthropic observed that experienced Claude Code users both approved more actions in advance and stopped more runs. The study tracks use of one product, not proof that giving AI more freedom is safe.
Nurses told WIRED that an AI-made schedule gave them unwanted shifts and a poor mix of skills. HCA says nursing leaders make final choices. The report does not prove that the tool harmed patients.
Microsoft researchers’ 2019 guidelines ask designers to help people spot and correct AI mistakes. The work was tested with design practitioners, not with today’s long-running AI tools.
NIST’s voluntary AI risk playbook groups work into four steps: govern, map, measure, and manage. For one small task, name an owner, a check, and what to do if the check fails.