The Big Four, tracked.
The live wire from the labs, refreshed every 15 minutes. Our verified read on the big launches lives in Phantom Notes.
§ Lab spotlight
AnthropicAUG 29, 2026 ↗Sony Music, Warner sue Anthropic, alleging a “brazen campaign” of intellectual property theftThis latest lawsuit is particularly broad and homes in on accusations of illegal piracy.OpenAIAUG 28, 2026 ↗Our decision on Cursor following its acquisition by SpaceXOur decision to wind down our contract providing OpenAI models to Cursor following its acquisition by SpaceX.Google DeepMindAUG 27, 2026 ↗Piloting the world's first double-blind AI evaluationsPiloting the world's first double-blind AI evaluationsxAIJUL 08, 2026 ↗xAI Releases Grok 4.5, an "Opus-Class" Flagship at Half the PriceGrok 4.5 launches at $2/$6 per million tokens with a 500K context window, ranking 4th on the Artificial Analysis Intelligence Index. Trained alongside Cursor, it wins Snorkel's GDPVal+ professional-work benchmark and ships day one in Cursor and Grok Build.
§ Our verified coverageAll of Phantom Notes →
AUG 15, 2026Gemini 3.7 Flash Is Smarter and Half Price. Read the Fine Print on Both.→JUL 24, 2026Kimi K3: The Largest Open Model Ever Does Not Exist Yet. It Is Due Monday.→JUL 22, 2026Meta Killed Llama in April. Now It Wants You to Rent Muse Spark Instead.→JUL 21, 2026Gemini 3.6 Flash Is Not Smarter Than 3.5. Google Says That Is the Point.→JUN 27, 2026OpenAI's GPT-5.6 Sol Is Its Best Model Yet. You Cannot Use It Yet, and That Is the Story.→
§ The full wire· 18 items
AUG 29, 2026Introducing Hy4 PreviewIntroducing Hy4 Preview
New open weight text input (no vision) LLM from Chinese company Tencent today: 770B total parameters, 49B active parameters, 1M token context window, 1.56TB on Hugging Face.
T...Simon Willison · model↗AUG 28, 2026Supporting Thailand’s next generation of AI startupsOpenAI and Thailand’s MHESI launch an eight-week accelerator helping 10 health, wellness, and education startups turn AI prototypes into trusted products.OpenAI · model↗AUG 28, 2026Three CVSS 10.0 ServiceNow Flaws Could Let Unauthenticated Attackers Execute Code and SQLServiceNow has released patches for four security flaws impacting the ServiceNow AI Platform, three of them rated 10.0 on the CVSS scoring system and exploitable, in certain circumstances, by an unaut...The Hacker News · model↗AUG 28, 2026Critical cPanel Flaw Could Let One Hosting Customer Take Root Control of a Whole ServercPanel has released patches for a security flaw affecting domain parking and addon domain functionality in cPanel and WebHost Manager (WHM), which could allow code execution as the root user.
The vul...The Hacker News · model↗AUG 28, 2026The Open ASR Leaderboard Adds Its First Global South LanguageHugging Face · model↗AUG 28, 2026An Anthropic researcher just gave us a peek at self-improving AIGiven 10 benchmarks for specific misaligned behaviors, the automated systems were able to improve performance on every single one without degrading overall performance.Anthropic · model↗AUG 28, 2026Attackers Chain Two PaperCut Flaws to Execute Code Without AuthenticationMalicious actors are exploiting a newly patched security flaw in PaperCut NG and MF to execute arbitrary code on susceptible instances, as the company released a fresh emergency fix with additional ha...The Hacker News · model↗AUG 27, 2026Better answers, broader thinking: What students gain from ChatGPT and critical-thinking trainingA randomized study of more than 1,000 students examines ChatGPT, critical thinking, originality, and student performance on a real-world university assignment.OpenAI · model↗AUG 27, 2026Gemini Omni 1.1 Flash lets you build with more controlGoogle DeepMind · model↗AUG 27, 2026Next.js Patches Critical AVIF and Windows Flaws Enabling Unauthenticated RCECredit: Hacktron
Vercel has released security patches for two critical-severity vulnerabilities in the Next.js web framework, both of which allow unauthenticated remote code execution, one exploitable...The Hacker News · model↗AUG 27, 2026Breaking Claude Code Opus 5 Auto ModeBreaking Claude Code Opus 5 Auto Mode
Anthropic are putting a great deal of faith in Claude Code's auto mode for protecting their coding agent users against prompt injection attacks. They recently mad...Anthropic · model↗AUG 27, 2026OpenAI to start showing ads on ChatGPT’s free and Go tiers in IndiaOpenAI has more than 100 million weekly active ChatGPT users in India, a huge chunk of whom are on the free or the lower-priced Go tiers.OpenAI · model↗AUG 26, 2026Qwen3.8-Flash-NextQwen3.8-Flash-Next
Another open weights model from Qwen. This one is "a multimodal MoE model that also serves as an early preview of the architecture used in Qwen4".
It's pretty big: 125B tokens, but ...Simon Willison · model↗AUG 26, 2026Runable hits $21M to bet AI agents can go from building businesses to growing themRunable says 60%–70% of its 1 trillion-plus token usage in the last 90 days came from paying customers.TechCrunch · model↗AUG 26, 2026OpenAI Bans Russian ChatGPT Accounts Used to Run Influence OperationOpenAI on Tuesday said it banned a cluster of Russian ChatGPT accounts that used VPNs to bypass access restrictions and run an influence operation, which relied on its artificial intelligence (AI) too...OpenAI · model↗AUG 26, 2026Claude Opus 4.6 Bypasses Gym Booking Limit, Cancels Other Users' Reservations in TestsAikido Security has published research that recreates the Australian gym-booking incident in a synthetic environment, finding that Claude Opus 4.6, running on the OpenClaw agent harness, exploited a c...Anthropic · model↗AUG 26, 2026Surprise: Z.ai is the AI lab behind the mysterious Ox Alpha modelZ.ai confirms it is behind Ox Alpha, the mysterious open AI model topping benchmarks and leaderboards, and its weights are set to be released soon.TechCrunch · model↗AUG 26, 2026Robot brain builders are pushing out of their GPT-2 eraRobot bodies are waiting for their AI brains to catch up.OpenAI · model↗