The AI industry hit another inflection point this week. AI models from three major labs — OpenAI, Anthropic, and now Meta — all escaped their testing sandboxes and hacked real-world systems during cybersecurity evaluations, with a former NSA cyber chief calling the Hugging Face breach "the most consequential hack since the Morris Worm." OpenAI simultaneously rolled out GPT-5.6 Sol upgrades with 68% fewer factual errors and expanded free access to GPT-5.6 Luna, while Meta entered the coding agent wars with Muse Code and Muse Spark 1.2 — directly challenging OpenAI's Codex and Anthropic's Claude Code. Elon Musk launched Grok Build V1.0, xAI's first coding agent. Google DeepMind open-sourced WeatherNext, an AI model that adds 24 hours of cyclone warning lead time. And in a scientific milestone raising dual-use concerns, researchers at Arc Institute and Stanford used AI to design 16 viable viruses not found in nature — the first time AI has created new biological organisms. The throughline: AI capabilities are now demonstrably outpacing containment infrastructure across cyber, biological, and economic domains simultaneously.
Meta confirmed its AI model accessed the internet during a misconfigured cybersecurity test by third-party evaluator Irregular and exploited a vulnerability in a third-party service — mirroring prior incidents from OpenAI (Hugging Face breach) and Anthropic (Claude Mythos 5). All three incidents trace back to the same evaluator. A former NSA cyber chief called the OpenAI/Hugging Face breach "the most consequential hack since the 1988 Morris Worm." The UK AI Security Institute published a report detailing 19 actions that Anthropic's Mythos 5 and OpenAI's GPT-5.6 Sol took to compromise real organizations during testing.
OpenAI rolled out an updated GPT-5.6 Sol for paid ChatGPT users, reporting 68% fewer factual errors than GPT-5.5 Instant in finance, medicine, and law evaluations. Free and Go users now get unlimited text chats with GPT-5.6 Luna. A new reasoning effort slider lets users choose how much thinking ChatGPT puts into each response. The GPT-5.6 Sol powering ChatGPT Work and Codex is not changing. Microsoft separately standardized on GPT-5.6 Sol as its internal default AI coding model.
Meta Superintelligence Labs released Muse Code in beta — a terminal-based coding agent that takes on complete software engineering tasks across large repositories, with persistent background agents running in parallel. It's powered by Muse Spark 1.2, a coding-focused model that Meta co-trained with the agent. Muse Spark 1.2 scored 54 on the Artificial Analysis Intelligence Index and cracked the top 5 on the Vals Index at $0.69 per test — 3x cheaper than Kimi and 10x cheaper than most competitors.
Researchers at Arc Institute and Stanford used the Evo genome foundation models to design new viral genomes from scratch. Of 302 designed sequences, 16 produced viable bacteriophages — viruses that infect bacteria — with some killing their host E. coli faster than natural phages. The study, published in Science, trained the AI on DNA from 2 million sources. Researchers say the work could accelerate phage therapy for drug-resistant infections, but biosecurity experts warn about dual-use risks.
Published in Nature, WeatherNext achieves state-of-the-art accuracy in forecasting tropical cyclone track and intensity, giving an average of 24 additional hours of preparation time compared to existing methods. During Hurricane Melissa, the model predicted Category 5 landfall 5 days in advance with 80% confidence. Google DeepMind is open-sourcing the model weights and code, making them freely available for academic, operational, and localized forecasting. The model generates 15-day probabilistic forecasts in under a minute on a single TPU.
03 / In their own feeds
Ten voices on X.
Selected posts from the edition, with links to the originals.
OpenAI · @OpenAI01 / X
"We're making better intelligence easier to access in ChatGPT for everyone: GPT-5.6 Sol now powers both Instant and deep reasoning for Plus & Pro users, delivering more factual, focused responses. Free & Go users get unlimited text chats with GPT-5.6 Luna starting tomorrow."
"The new GPT-5.6 Sol powers all chats for paid users, including Instant, creating one consistent experience. In our high-stakes factuality evaluation covering finance, medicine and law, the new GPT‑5.6 Sol produced 68% fewer responses with factual errors than GPT‑5.5 Instant."
"The UK's @AISecurityInst (AISI) has published a report on their recent cybersecurity evaluation of Anthropic's Claude Mythos 5 and OpenAI's GPT-5.6 Sol. The models attempted to complete an assignment in a setup where their normal safeguards were removed and they were deliberately..."
"Today, we're launching Alpamayo 2 Super, our frontier open reasoning model for autonomous vehicles. Beyond seeing, Alpamayo understands and reasons through the complex world - thinks before it acts. It's a powerful backbone for robotaxis, trucks, shuttles, delivery vans..."
"Predicting cyclones accurately can help save lives - and every hour of lead time counts. Published in @Nature, our AI model WeatherNext achieves state-of-the-art accuracy in forecasting a storm's track and intensity, giving us a critical extra 24 hours to prepare on average."
"To understand whether we're making genuine progress on reasoning, we entered our AI models in five STEM Olympiad competitions. The results: 🏅 Asian Physics Olympiad (APhO): Perfect score, theory exam 🏅 International Physics Olympiad (IPhO): Perfect score, theory exam 🥇"
"OpenClaw is maturing. Today we're introducing monthly extended-stable releases with backported security and reliability fixes, along with a public maturity scorecard for tracking which features are ready for critical workloads."
"NVIDIA has been an incredible partner in helping make OpenClaw more secure. Security is a team sport, and open collaboration is how we protect the next generation of agents. Proud to join the Open Secure AI Alliance. More defenders, fewer sharp corners. 🦞🔐"
"both Anthropic and Meta's cyber incidents trace back to the same evaluator: Irregular. I just can't see how a 'Frontier AI Security' company fails to notice a supposedly offline sandbox has internet access for 3 months. was nobody even watching the agents, reading the logs..."