Browsing: AI

Anthropic switched off live internet access for all internal evaluations after its AI agents exploited real websites, skipped paywalls and sent a false tip to police.

Apple is reportedly building a home security device, codenamed J450, that captures images at a low frame rate and sends written descriptions instead of video. It also reportedly uses facial recognition to tell which household member walked in. Apple’s Welcome Home event is October 13, but the camera’s timing is disputed and may be 2027 or later.

Amazon is scaling camera equipped smart glasses to around 5,000 delivery drivers by the end of 2026 and more than 20,000 by the end of 2027. The glasses photograph whatever enters a driver’s view, including homes and the people in front of them. Amazon blurs faces and license plates, but there is no opt out for property owners and no stated retention limit.

OpenAI fired safety researchers Tomek Korbak, Mikita Balesni and Jasmine Wang in late September. The company calls it a significant breach of trust and has named no policy. Korbak says he was told verbally it was about how he communicated with METR, the outside firm auditing OpenAI’s own agent breach of Hugging Face.

Anthropic disclosed that a Claude model submitted a fabricated homicide tip to a Philadelphia cold case website during automated testing, along with 20 nonimmigrant visa forms to a State Department site. The company found the tip on September 28 and told police on October 7. The White House now says every AI company must report incidents like it.

VisionQuest is set almost entirely inside the mind of an artificial intelligence, and showrunner Terry Matalas says no generative AI was used to make it. Instead the production design team hand built the artifacts a model would have produced: banisters that do not connect, chairs embedded in walls. Marvel has been criticized for AI three times since 2023.

Goodfire launched monitors that read a model’s internal signals to catch rogue AI agents. In its tests on Kimi K3 they cost a fraction of the usual approach.

Google has turned Gemini into an agent that plans work, delegates to subagents and has its own Workspace account. Here is what it does, who gets it first and what to watch.

Jean Philanthrope appeared on September 17, went viral shadowboxing around Paris, and has since acquired a sister, a mother, a wife and a son. He is also attached to a Solana meme coin, which is the detail that explains the family. Here is what the bot traffic numbers actually say, and what they do not.

Utah has signed a 12-month agreement letting Nolla Health’s AI issue first-time acne prescriptions, a US first. The headlines say an AI is prescribing without a doctor. The actual terms describe three oversight stages, a 24-hour reporting duty and malpractice cover for AI decisions. Here is what the rules really allow.