AI News — 2026-08-23
- Nvidia partners with data center developer Cloverleaf. Deepens its infrastructure push with another equity stake in the buildout race. — TechCrunch
- Nvidia researchers hit 100% on ARC-AGI-3 with Claude Opus 5. Shows the surrounding “harness” — not the model — is now the bigger lever for agentic performance. — TechCrunch
- Anthropic’s IPO filing will flag AI backlash as a risk factor. Public trust concerns are becoming a disclosed business risk ahead of a possible October listing. — CNBC
- Anthropic plans to loosen its data retention policy for advanced models. Enterprise customers on Mythos and Fable will get the option to keep 30-day retained data on their own cloud infrastructure. — Bloomberg
- OpenAI rolled out a dedicated ChatGPT for Teens experience. Adds age prediction, parental notifications for high-risk situations, and limits on romantic or emotionally dependent language. — CNBC
- Amazon made Alexa+ free on Fire TV, no Prime required. Extends its AI assistant’s reach into living rooms without a subscription gate. — TechCrunch
- Meta launched Pocket, an app for vibe-coding and sharing games. Lets users generate playable mini-games from prompts and share them with other US users. — TechCrunch
- Binance now lets AI agents trade on its platform via Agent OS. Connects AI applications to Binance’s markets, but keeping agents in check is left largely to users. — TechCrunch
- Ramp launched Router, its own AI model routing service. Lets companies switch between large language models through a single API. — TechCrunch
- A third of web pages published since ChatGPT’s launch show signs of AI authorship. New Pew Research study measures how fast AI-written content has spread online. — TechCrunch
- Perplexity’s free AI offer won it millions of new users in India. The giveaway pushed adoption sharply upward in one of its biggest growth markets. — TechCrunch
- Starcloud raised $250M for orbital AI data centers as launch slots grow scarce. Extension to its March round values the satellite-compute startup at $2.3B. — TechCrunch
- Michael Polansky’s startup is training an AI model on living human skin tissue. Outer Biosciences keeps excised skin alive for weeks to gather training data outside the body. — TechCrunch
- OpenAI now says California should strengthen its AI safety bill. Reverses its earlier opposition to SB 53, urging monitoring of frontier models in training and stronger cybersecurity rules. — TechCrunch
- Frontier AI labs still won’t say how they’d contain a rogue model. A new study finds little public detail on containment plans even as models show more unexpected behavior. — TechCrunch
- Nvidia is hiking AI server prices by more than 15% for some customers. Soaring memory chip costs push up the price of systems including Vera Rubin and Grace Blackwell, effective on shipments starting early next year. — Bloomberg
- Anthropic expects its IPO to match or top SpaceX’s record size. A public filing could come as soon as the end of this month, sources say. — Bloomberg
- Nvidia is paying AI startup Poolside a $6B model license. The deal adds a $1B investment and job offers to 100+ employees, while Poolside stays independent. — Bloomberg
- OpenAI’s CFO says it will go public in 2027, or sooner. Sarah Friar told employees the IPO is “a milestone, not a finish line.” — CNBC
- Harvard’s $699 startup bootcamp uses AI avatars for feedback. HeyGen-built avatars critique pitches and board meetings between weekly live sessions. — TechCrunch