1. Project Glasswing — Anthropic’s unreleased Claude Mythos Preview autonomously found thousands of zero-day security vulnerabilities across every major OS and browser, including a 27-year-old OpenBSD bug — withheld from public release and limited to 12 launch partners including Apple, Google, Microsoft, NVIDIA, and AWS.
2. Anthropic Revenue Surge — Anthropic’s run-rate revenue tripled to $30 billion since January, with 1,000+ enterprise customers paying over $1M annually, and secured 3.5GW of Google TPU capacity through 2027 via a new Broadcom deal.
3. Altman Investigation — The New Yorker’s Ronan Farrow published a sweeping investigation drawing on 100+ interviews and Ilya Sutskever’s internal memos, with a senior Microsoft executive comparing Altman to “Bernie Madoff.”
4. OpenAI Social Contract — OpenAI released a 13-page blueprint proposing robot taxes, sovereign wealth funds paying dividends to all Americans, 4-day workweeks, and a guaranteed “Right to AI” access enshrined in law.
5. GLM-5.1 Breakthrough — Chinese lab Z AI open-sourced a 744B MoE model that topped GPT-5.4 and Opus 4.6 on SWE-Bench Pro with a score of 58.4, capable of 8-hour autonomous coding sessions under MIT license.
6. Google Gemma 4 — Google released four Gemma 4 models under Apache 2.0 for the first time, removing commercial restrictions that had previously pushed enterprises toward competitors — smallest variants run offline on phones.
7. Anti-Distillation Pact — OpenAI, Anthropic, and Google began sharing threat intelligence through the Frontier Model Forum to detect and block Chinese AI labs from distilling their models; Anthropic documented 16 million adversarial exchanges from DeepSeek, Moonshot AI, and MiniMax.
8. Meta AI Models — Meta is launching the first models from Alexandr Wang’s Superintelligence Labs with a hybrid open/closed strategy, after Llama 4 failed to keep pace with competitors.
9. Solo $1.8B Company — Matthew Gallagher built Medvi, a GLP-1 telehealth company, from $20K to $1.8B in projected 2026 revenue using only AI tools and a two-person team — validating Sam Altman’s prediction of the one-person billion-dollar company.
10. Anthropic Blocks Agents — Anthropic cut off flat-rate subscriptions from third-party agent platforms like OpenClaw, with some users facing up to 50x cost increases as autonomous agent usage overwhelmed subscription pricing.
Alibaba Qwen3.6-Plus — Flagship reasoning model with always-on thinking mode, 1M-token context, and multimodal support; rivals Opus 4.5 on coding benchmarks and supports agentic workflows via Claude Code and OpenClaw.
OpenAI Acquires TBPN — OpenAI’s first media acquisition: the live tech talk show TBPN for reportedly hundreds of millions, with full editorial independence; the show generated $5M in ad revenue in 2025 and is on track for $30M+ this year.
Cursor 3 — Rebuilt agent-first interface letting developers orchestrate parallel fleets of local and cloud coding agents across repos, environments, and remote SSH — with seamless local-to-cloud handoff.
Microsoft MAI-Transcribe-1 — New speech-to-text model topping accuracy benchmarks across 25 languages at $0.36/audio hour, outperforming Whisper and Gemini Flash; now in public preview on Microsoft Foundry.
Netflix VOID — Open-source Apache 2.0 video object removal framework that erases objects while realistically rewriting the physics of the affected scene; preferred by 65% of evaluators over Runway, DiffuEraser, and other tools.
Mystery Model — “HappyHorse-1.0” debuted at No. 1 on video generation leaderboards, surpassing ByteDance’s Seedance 2.0 by nearly 60 Elo points before mysteriously disappearing from the rankings.
Intel Partnership — Intel joined Elon Musk’s Terafab project alongside SpaceX, Tesla, and xAI to build a greenfield semiconductor fab in Texas targeting 1 TW/year of AI compute production.
Microsoft Open Source — Bing team released Harrier, a 27B embedding model ranking #1 on multilingual MTEB v2, supporting 100+ languages with a 32K context window; available on Hugging Face under MIT license.
Claude Code Leaks, Part 1 — Breaking down what Anthropic accidentally exposed: 1,900 files, 500K+ lines of Claude Code’s source, and unreleased features hiding inside.
My First Claude Code Demo — I built a /skill that sends me a morning brief with talking points for every meeting on my calendar that day — pulling from Gmail and Google Calendar.
Our Future with AI — My take: as AI handles more of our screen time, we might actually end up more present with each other. I’m genuinely excited about that possibility.
No posts
Comments
Nothing yet. Say the first thing.
Sign in to join the conversation.