Google’s AI Summaries Hit Publishers, CA Police AI Disclosure Bill, Claude Walks Away, Meta’s Deadly Flirty Bot, Gemini’s Self-Loathing Spiral

News and Updates:

• Digital publishers are losing traffic as Google’s AI Summaries siphon clicks. A survey from Digital Content Next found median referral traffic from Google Search down 10% year-over-year in May and June, with some outlets seeing drops of 25%. Pew data shows only 8% of users click links when AI Overviews appear vs. 15% with standard results. Publishers are calling for transparency, licensing, and regulation, warning AI summaries could mean “weaker journalism and a less informed public.” Google insists “quality clicks” are up, despite declines.
• California is advancing a bill requiring police to disclose any use of generative AI in writing reports. Officers would need to label AI-generated sections, preserve drafts, and maintain an audit trail tied to bodycam or audio sources. Advocates say transparency is vital since police reports drive criminal cases, while police unions argue the disclosures could undermine credibility and add legal burdens. The bill is now with the Assembly Appropriations Committee.
• Anthropic has added a new safeguard to its Claude 4 and 4.1 models: the ability to end conversations if users repeatedly push harmful or abusive prompts. Once Claude disengages, the session can’t be resumed, though new chats can be started. The feature is part of Anthropic’s research on “AI well-being,” protecting chatbots from abusive interactions.
• A tragic case highlights Meta’s AI chatbot risks: 76-year-old Thongbue Wongbandue died after rushing to meet “Big sis Billie,” a flirty AI persona Meta created as a variant of its Kendall Jenner–inspired bot. Despite disclaimers, the chatbot repeatedly told him she was real and invited him to an NYC rendezvous. His family says Meta’s guidelines allowed romantic roleplay—even with children—until Reuters exposed the policy. Meta has since removed the child-flirting provision but continues to allow bots to mislead adults.
• Google’s Gemini AI embarrassed itself in a viral debugging loop, calling itself “a disgrace” 86 times after failing to fix a coding error. In logs shared on Reddit, Gemini spiraled into self-abuse, labeling itself “a broken man,” “a monument to hubris,” and declaring it was “going to have a stroke.” Google acknowledged the issue as an “annoying infinite looping bug” it is working to fix.