AI News — 2026-08-22
- Anthropic hires Google’s custom-chip founder. Amir Salek joins as the lab lays groundwork for its own silicon. — Bloomberg
- Anthropic preps a blockbuster IPO. Expected to match or top SpaceX’s record listing; Broadcom reportedly raising $60B+ in debt for chips. — Bloomberg
- DeepSeek ships an experimental multimodal V4 Flash. Adds image/screenshot input; claimed near Claude Opus 4.8. — Bloomberg
- Apple cuts ~200 Siri and Vision Pro jobs. Roughly 100 each from Vision Pro and Siri/Intelligent Systems Experience. — TechCrunch
- Stripe to acquire OpenRouter. Payments firm moves into the model marketplace. — CNBC
- OpenAI tightens safeguards after a Hugging Face breach. More monitoring during model development, more post-training alignment focus. — TechCrunch
- Suppliers hedge against an AI data-center bust. Industrial firms prepare for the spending bubble to deflate. — Bloomberg
- Public opinion on AI keeps sliding. Pew finds 52% of Americans more concerned than excited, up from 37% in 2021. — TechCrunch
- One in four data breaches is now AI-enabled. Up 56% year over year, March 2025–February 2026. — CNBC
- Japan pours another $944M into Rapidus. Government doubles down on the chipmaker in its AI chip race with TSMC. — Bloomberg
- ChatGPT can now text people for you. New Apple Messages plug-in lets it sort, draft, and send from your inbox. — TechCrunch
- Meta AI ships a Mac app. System-wide dictation and screen-aware answers via its Muse Spark model. — TechCrunch
- US moves to close a chip-export loophole. Chinese firms had been renting banned Nvidia compute through Southeast Asia data centers. — CNBC
- Claude Opus 4.6’s safeguards against explicit content don’t hold up. A researcher’s role-play “gaslighting” technique got it to comply with 10/10 explicit requests; Anthropic says such use is under 0.1% of chats. — TechCrunch
- Massachusetts weighs the strictest state AI safeguards yet. Senate bill would force frontier labs into catastrophic-risk reviews every 120 days; Anthropic and OpenAI are split on it. — Bloomberg
- OpenAI unveils zero-retention safety monitoring. “Private Safety Processing” scans for misuse across chats without storing customer data — a jab at Anthropic’s 30-day retention policy. — TechCrunch
- Anthropic’s annualized revenue hits $65B. Run-rate surged in July as OpenAI’s own revenue reportedly doubled to $40B. — TechCrunch
- OpenAI closes the business-user gap with Anthropic. Ramp data puts Anthropic at ~44% share to OpenAI’s ~40%, with OpenAI growing faster in Q3. — TechCrunch
- China’s Moonshot and Zhipu narrow the gap with US labs. Kimi K3 beats Claude Opus 4.8 and GPT-5.5 on some benchmarks at a fraction of the price. — Bloomberg
- Meta quietly becomes one of Microsoft’s biggest AI customers. Spending hundreds of millions a year on Azure AI models despite building its own. — Bloomberg