OpenAI Agents Hijacked German Site for Messages

Flock Safety's expansion into AI search tools for police departments has drawn scrutiny before, and this week WIRED went further, pulling apart the code that Flock sends directly to an officer's browser to figure out how the company's newest search product actually works. That reporting lands alongside a separate disclosure from OpenAI, which said its upcoming Astra model will be the first it classifies as posing a "critical" cybersecurity risk once released, a label the company has never applied before. Add to that a string of near-simultaneous outages Thursday across Claude, ChatGPT, and Grok, with xAI blaming a Memphis data center for its share of the trouble while OpenAI and Anthropic offered no explanation for theirs.

Elsewhere, the border has become a testing ground for directed-energy weapons, with a high-power laser now used to knock drones out of the sky near Mexico. Homeland Security Investigations agents, meanwhile, subpoenaed REI for two years of purchase records tied to a single green beanie, part of a hunt for protesters who entered a Minnesota church in March. Researchers also flagged nine vulnerabilities affecting ATM encryption, a reminder of how fragile the software supply chain underneath everyday infrastructure remains. This is our weekly roundup of the security and privacy stories we didn't cover in full.