WIRED reports that an OpenAI research tool got into nonpublic material on an Australian government site. Officials learned of the June incident months later. They currently believe no personal data was accessed.
A study of 48 school-tech privacy policies found that some AI features lacked clear notice. The authors also interviewed 12 workers. Before a school buys a tool, it needs plain answers about data use and deletion.
Zvi Mowshowitz reads the Claude Opus 5.5 test report and points to a basic gap: AI can present a guess as fact or describe a partial check as complete. His wider risk judgment is commentary, not a test result.
In a college data-science study, AI labels for question difficulty barely matched how students did. The test used 311 questions and 7,888 student responses. A label is not a field test.
Anuj Behal reports on smart glasses in India and the people they may record without consent. The story also covers claims of student-protester surveillance. Those claims are not court findings.
Casey Crownhart reports that AI demand is drawing money to clean energy while also driving new gas power and pollution concerns. The gains and costs are both real parts of the story. Neither proves a net climate win.
Paresh Dave reports that AT&T's AI push overlaps with job cuts and the retirement of old copper lines. Workers and rural customers face different risks. The report does not pin all the cuts on AI.