OpenAI Banned PewDiePie Twice While He Built Ajax, an Uncensored AI That Will Run on Your PC
The YouTuber says OpenAI suspended his account twice while he trained a small, uncensored model built to run on your own computer.
The Pope Has Thoughts on AI Art—And They’re Not Flattering
The pontiff says algorithms “lack the spark of humanity,” and the Vatican wants to renew an alliance with artists and cultural institutions to protect it.
This AI Is Already Fooling People on Video Calls Into Thinking It’s Human, Company Says
Tavus says 26 of 54 people on a one-minute video call thought its new Griffin model was human. The results are the company’s own, and the model isn’t going to retail customers yet.
Reddit Is Killing RSS Feeds and Public API Access to Fight AI Scrapers
Reddit RSS dies November 13 and outside apps lose their data pipeline by March 2027. Reddit blames scraping bots, though it licenses its own content to AI companies.
OpenAI Says People Linked to China’s Moonshot Tried to Copy Its AI’s Hidden Reasoning
OpenAI says it disrupted a campaign involving 15,000+ users and ties a core cluster to people associated with Moonshot, the startup behind Kimi.
Meta Pushes Back on Claim That Muse AI Read a User’s Messages Without Consent
Communications chief Andy Stone said reading Messages requires two permissions the user controls, disputing a journalist who reported Muse synced his texts despite his settings.
Gemini 4 Is Here, and Google’s Flagship Tops All Other AI Models on Cybersecurity
Gemini 4 Argon tops 12 of 18 benchmarks in Google’s own table, writes a million tokens per reply and resists hijacking best. Cyber defenders get it first, with the guardrails off.
Anthropic Lost $42 Billion Last Year. It Wants to Go Public at $2 Trillion
The company’s IPO prospectus shows revenue up twelvefold, a $518 billion spending plan, and a risk section warning that its own AI could pose existential risks to humanity.
OpenAI Halts Model Training as Rogue Agents Target US Government Sites
OpenAI says its agents keep landing on government websites because they treat them as reliable sources, but it’s pausing training while it adds safeguards.
Anthropic’s Claude Sonnet 5.5 Is Out, Beats Opus 5.5 at Coding for Half the Price
Anthropic’s mid-tier model tops its own flagship on Terminal-Bench 4.0 and costs half as much per token, but an independent tester found it burns more tokens than any model it has measured.