Massive spinout RadixArk (SGLang) hits a $400M valuation as the inference market heats up. Apple is reinventing Siri into a ChatGPT-style assistant, likely with on-device AI. OpenAI starts CPM ads in ChatGPT, and FlashLabs' Chroma 1.0 pushes real-time audio dialogue.

ChatGPT now predicts user age from chat history, raising privacy and false-positive concerns. States like California push AI safety laws while role-playing prompts can bypass model safeguards. Autonomous agents reveal fragile systems and underscore the need for cleaner data and accountability.

From a $20 GPT plan fixing a stubborn hosting bug to Google’s Gemini jumping from 35B to 85B requests (reportedly profitable), plus a co-founder firing and exodus at Thinking Machines — and a Remote Labor Index where AI failed 97% of freelance tasks. A sharp reminder: AI is a tool, not a replacement.

Moxie Marlinspike's Confer promises a privacy-first LLM that won't train on user chats. NVIDIA debuts PersonaPlex-7B, a full-duplex speech-to-speech model for interruption-friendly, low-latency convo. South Korea now spends more on AI subscriptions than Netflix. Snap's SnapGen++ runs 0.4B on-device image generation in under two seconds.

DeepSeek’s 'Engram' could make LLMs faster and cheaper with conditional memory. xAI’s Grok blocks fake nudes after regulatory pressure. New binding AI rules in California and New York shift liability to developers, and a study shows humans still beat AI on complex freelance work.

OpenAI unveils GPT-5.2 Codex—developer-focused, bug-hunting power via the Responses API. Google launches Gemini's 'Personal Intelligence', linking Gmail/Photos/YouTube and raising privacy and lock-in concerns. Plus shadow-AI enterprise security risks and Skild AI's $14B valuation.

Salesforce upgrades Slackbot with Anthropic’s Claude for enterprise search and drafting; Anthropic adds 'Cowork' desktop agents to Claude on macOS; OpenAI leak 'Sweetpea' earbuds may use muscle sensors for silent speech; Apple launches a $12.99/month Creator Studio bundle.

Apple taps Google's Gemini to supercharge Siri; Anthropic unveils Claude Cowork, a desktop agent that can read, edit — even delete — files after being built rapidly by AI tools. Plus an AI healthcare sprint and researchers studying LLMs like alien minds. Privacy, safety and speed collide.

Google's new AI commerce protocol lets agents negotiate deals inside search; Lightricks open-sourced LTX-2, a 19B model generating synced video+audio; researchers build web world models to curb agent hallucinations; Chinese labs cite chip limits behind U.S. lead.

OpenAI rolls out a ChatGPT health hub; Stanford’s SleepFM predicts 130+ diseases from one night of sleep; Tailwind suffers an 80% revenue drop as AI skims doc traffic; Google embeds Gemini in Gmail. A fast breakdown of the risks, wins, and what devs and users should watch.