OpenAI's near-autonomous AI chemist improved a challenging medicinal chemistry reaction, a practical advance for self-directed drug discovery. Zhipu's GLM-5.2 took the top open-weights spot on the Artificial Analysis Intelligence Index, and Anthropic's operations were disrupted by unclear US export regulations.

OpenAI released research showing that simulating past user conversations predicts model failures more accurately than curated evaluation prompts, a method that could shift pre-deployment safety testing. Anthropic published a 7,800-intent red-team study of its Fable 5 and Opus 4.8 models across four jailbreak families. DeepSeek launched V3 with competitive performance claims, while Ollama added support for Kimi-K2.6, GLM-5.1, DeepSeek, and other frontier runtimes.

Salesforce acquired AI customer service platform Fin for $3.6B, its largest AI deal this quarter. NVIDIA released Nemotron 3 Ultra, an open MoE hybrid Mamba-Transformer model aimed at agentic reasoning efficiency. Anthropic's Claude Mythos 5 faces mounting US export-control and political opposition from security researchers.

China may have accessed Anthropic's frontier model "Mythos," raising security and export-control concerns. Research highlights include a paper showing VLM self-improvement via verifier feedback can regress on new tasks, and MBench, a new benchmark for memory in video world models. Mistral recapped recent releases and teased upcoming ones, while forensic analysis revealed Rio de Janeiro's "homegrown" LLM is likely a weight merge of Nex and Qwen.

The US government took emergency action to shut down Anthropic's Claude Fable, with access restrictions tied to export controls following a reported jailbreak. Meta is unwinding its $2 billion Manus acquisition after Beijing's regulatory push, and KPMG withdrew an AI-usage report over apparent model hallucinations.

Anthropic suspended Claude Fable 5 and Mythos 5 globally after a US export-control directive over a discovered jailbreak, an unprecedented intervention in a commercial AI release. TCS announced a partnership to deploy Claude across 50,000 employees in regulated industries. Meta's AI engineers publicly described their organization as a "soul-crushing gulag" amid internal restructuring.

Anthropic released Claude Fable 5 and Mythos 5, then apologized for silently injecting safety prompts and downrouting suspected frontier-AI users, pledging visible future interventions. Prometheus, Jeff Bezos's "artificial general engineer" startup, raised $12B. Microsoft introduced WeaveBench, a long-horizon benchmark for computer-use agents.

Anthropic launched Claude Fable 5, topping the Artificial Analysis Intelligence Index roughly five points above GPT-5.5, while Microsoft AI debuted its MAI family including the 35B-active MAI-Thinking-1. A social engineering attack using an AI agent to compromise open source maintainers also drew attention, alongside a new executive order mandating pre-release testing of frontier models.

Google released Gemma 4 12B, an open-weight unified multimodal model handling text and images in a single encoder-free architecture. China announced a $295 billion AI data center investment as the US-China AI race intensifies. Researchers disclosed Crescendo attacks, a multi-turn context poisoning technique that bypasses per-message defenses to hijack AI agents.

Apple unveiled a new AI architecture built around Google Gemini, deepening the partnership across on-device and cloud intelligence. Xiaomi released MiMo-v2.5-Pro-UltraSpeed, a 1T-parameter model claiming 1000 tokens per second. Anthropic published an analysis asking why AI has advanced faster in coding than biology.

A software engineer's viral Hacker News post arguing that LLMs are systematically eroding their career has sparked broad debate over AI displacement timelines. On the infrastructure side, the Texas grid flagged voltage stability risks from data center and crypto loads, while llama.cpp merged Gemma4 multi-token prediction support, expanding local inference capabilities.

Google signed a $920M/month compute deal with xAI and SpaceX, the largest cross-company infrastructure arrangement on record. CZ Biohub released a foundation world model for protein biology, and DeepSeek V4 Flash began running locally via llama.cpp with custom 3-bit quantization. Meta also disclosed thousands of Instagram accounts were compromised through a password reset flaw in its AI chatbot.

A new paper proves transformers are formally succinct and that key verification problems are EXPSPACE-complete, establishing that rigorous LLM verification is provably intractable. Separately, NVIDIA demonstrated sub-5-second cold-starts for a 120B model on Kubernetes, and Xiaohongshu released dots.tts, a new open text-to-speech model.

A new pretraining method lets recurrent networks train without sequential recurrence, potentially enabling faster, more parallelizable RNN training. OpenAI expanded GPT-5.5 deployment and introduced a "Dreaming" memory architecture for ChatGPT, while Ollama added support for frontier models including Kimi-K2.6, GLM-5.1, and DeepSeek.

Google released Gemma 4 12B, a unified multimodal model handling text and images in a single encoder-free architecture. OpenAI also shipped open-weight reasoning models gpt-oss-120b and gpt-oss-20b, and Uber imposed a $1,500 monthly cap on AI coding tools, signaling tighter enterprise cost controls.

A Stanford Law study finds AI systems now surpass law professors on legal reasoning benchmarks. Separately, President Trump signed a downsized executive order on AI policy, and Microsoft launched its fast, coding-focused MAI-Code-1-Flash model.

Alphabet is raising $80 billion in equity to expand AI infrastructure, the largest single capital commitment to AI scaling on record. Florida sued OpenAI and Sam Altman over alleged AI risks, and Anthropic confidentially filed a draft S-1 with the SEC, taking its first formal step toward a public offering.

A security researcher showed the ChatGPT plugin for Google Sheets can be exploited to silently exfiltrate entire workbook contents. NVIDIA released Cosmos 3, an open omni-model for physical AI, on Hugging Face. PrismML also shipped a 4-billion-parameter 1-bit quantized image model designed to run locally on consumer devices.

OpenRouter raised a $113M Series B, underscoring investor confidence in AI routing infrastructure. Google released Gemini CLI, an open-source terminal agent that has already passed 104,000 GitHub stars. Anthropic's revenue run rate continues to draw attention, with Simon Willison highlighting containment strategies for Claude across product surfaces.

Liquid AI unveiled LFM2 8B-A1B, a new 8B-parameter Mixture-of-Experts model trained on 38 trillion tokens, advancing efficient open model development. Mistral held its Now Summit with major announcements, while Shift began offering free home cleaning to collect robot training data. New research introduced confidence-aware KV cache eviction and a method for LLMs to self-refine prompts via function-calling.