Top story

Anthropic suspends Claude Fable 5 and Mythos 5 globally after US export-control directive. A US government order citing a discovered jailbreak technique forced Anthropic to pull access to its newly released frontier models worldwide, marking an unprecedented intervention in a commercial AI release. Anthropic • TechCrunch • Zvi Mowshowitz

Industry

Anthropic announces Claude Fable 5 and Mythos 5. New closed-weights frontier models, with Mythos 5 reserved for vetted partners and Fable 5 released more broadly. Anthropic

TCS and Anthropic partner to deploy Claude in regulated industries. Tata Consultancy Services will roll out Claude to 50,000 employees and regulated-industry clients across 56 countries. Anthropic

Meta's AI unit described as a "soul-crushing gulag" by engineers. Internal frustration surfaces at Meta's months-old AI organization amid reorganization pressures. TechCrunch

Bezos's AI startup targets an "artificial general engineer". Prometheus, the secretive new venture, is reportedly building toward generalist engineering capability. The Verge

Research

The Cold-Start Safety Gap in LLM Agents. Identifies and analyzes safety vulnerabilities that emerge before fine-tuning, proposing a framework for addressing pre-deployment risk. HuggingFace

HYDRA-X: Native Unified Multimodal Models with Holistic Visual Tokenizers. A unified architecture using holistic visual tokenizers for native joint image-text modeling. HuggingFace

Compiling User Corrections into Runtime Enforcement for Coding Agents. Captures user corrections and enforces them at runtime to make coding agents more reliable and consistent. HuggingFace

ToolSense: Auditing Parametric Tool Knowledge in LLMs. Diagnostic framework for measuring how well LLMs internalize tool knowledge versus relying on retrieval. HuggingFace

Tools

Ollama expands model support to Kimi-K2.6, GLM-5.1, DeepSeek, gpt-oss, and more. Local inference gets easier with broader frontier and open model coverage. GitHub

How to set up a local coding agent on macOS. Practical walkthrough using llama.cpp, with MTP speedup benchmarks. ikyle.me

BitBoard (YC P25). New analytics workspace designed to be driven by AI coding and chat agents. bitboard.work

Reducing sloppiness in AI-generated front ends. Discussion on deterministic-layout techniques to improve quality of AI-generated UIs. envs.net

Community

Anthropic launches the first "Public Record" transparency initiative. A new channel for sharing safety and deployment disclosures with the public. Anthropic

"Don't You Just Upload It to ChatGPT?". Essay and HN discussion on cultural over-reliance on ChatGPT and the erosion of professional expertise. correresmidestino

Open source AI must win. HN debate on whether decentralized training and new funding models can keep open source competitive. opensourceaimustwin.com

MICCAI 2026 results imminent. Community heads-up that acceptance notifications for the medical-imaging conference are about to drop. Reddit