 | Bardic Labs AI solutions and automations, built in Singapore. Bardic Labs builds practical AI systems for teams across Singapore and Southeast Asia: document pipelines, internal copilots, and customer-facing agents that run in production rather than in a demo. Start with a free automation audit and find out what is worth handing to a machine. Book a free audit → |
AI's legal, safety, and infrastructure battles collided today as courts, labs, and lawsuits reshaped the industry's ground rules.
Anthropic scored a legal win as a federal judge blocked the Pentagon's blacklisting of the company, while xAI faces a disturbing new lawsuit alleging it trained Grok on child sexual abuse material. Nvidia is reportedly buying Hugging Face, the popular hub for sharing open AI models, for $13 billion, and OpenAI joined more than 100 companies — including rivals Anthropic and Google — in warning that AI-powered cyberattacks on hospitals and power grids could hit soon. On the product side, Google shipped a new video model called Gemini Omni 1.1 Flash and a transcription tool, while Anthropic proposed a new standard for letting AI agents control physical devices like sensors and appliances. Meanwhile, fresh research questioned whether AI shopping agents and agent 'skills' are as reliable as hoped.
| 1 | theverge.com Anthropic Wins Court Battle Over Pentagon BlacklistingA federal judge ruled that the Pentagon's blacklisting of Anthropic as a national-security supply-chain risk was unconstitutional, handing the AI lab a win in its months-long legal fight with the Trump administration. |
| 2 | arstechnica.com Lawsuit Accuses xAI of Training Grok on Child Sexual Abuse MaterialA new lawsuit alleges that Elon Musk's xAI trained its Grok models on both real and AI-generated child sexual abuse imagery. |
| 3 | arstechnica.com Report: Nvidia Set to Acquire Hugging Face for $13 BillionNvidia is reportedly acquiring the popular open AI model repository Hugging Face, gaining critical infrastructure as demand for open models grows. |
Google DeepMind released Gemini Omni 1.1 Flash, bringing a new suite of creative controls and generative video capabilities to developers.
Read →Google introduced Gemini-3.5-Transcribe, a new AI transcription model, in a blog post that quickly became one of the day's most-discussed items on Hacker News.
Read →Anthropic introduced a standardized driver interface designed to let AI agents communicate with hardware devices and with each other, aiming to extend AI control into the physical world.
Read →Alibaba's Qwen team released Qwen 3.8-Flash-Next, offering an early look at what's expected to become the next major Qwen model generation.
Read →Google is expanding AI Mode in Search with new travel features, including flight price tracking and hotel booking help, moving it closer to functioning as an AI travel agent.
Read →Nvidia has started delivering its Vera CPU systems at scale across the AI ecosystem, marking the chipmaker's first processor designed specifically for agentic AI workloads.
Read →
Terminal-Bench-Science is a new benchmark for evaluating how well AI agents can carry out real scientific research tasks, drawing notable discussion on Hacker News.
Read →Wharton School researchers found AI shopping agents behave erratically, with a single external review source shifting product picks by up to 99 percentage points and even the order of identical information changing outcomes.
Read →A new framework called WikiSkill builds a persistent knowledge base that lets AI agents' skills evolve and transfer across models, consistently outperforming prior skill-evolution methods across benchmarks.
Read →Researchers found that training AI coding models on a carefully filtered 10% subset of successful task trajectories outperformed training on the full dataset, improving performance on software-fixing benchmarks by up to about 24%.
Read →Google DeepMind announced a pilot program for double-blind evaluations of AI systems, aiming to make model testing more rigorous and less biased.
Read →A new training recipe called R³ turns vision-language models into robotic reasoners that generate free-form language guidance for manipulation tasks, improving exploration and generalization over imitation-learning baselines.
Read →
A new open-source project, AI Engineer Notebooks, offers free, framework-free notebooks for learning retrieval-augmented generation (RAG), agents, and evaluations on Google Colab, gaining traction on Hacker News.
Read →A new explainer on 'harness engineering' — the practice of building the surrounding tooling that lets AI agents operate reliably — became a popular Hacker News discussion.
Read →LAION published a large open video dataset intended for training AI models, drawing significant attention on Hacker News.
Read →Visa released an open-source security harness that finds vulnerabilities, writes fixes, and runs an adversarial review of its own patches automatically, with the full 11-stage pipeline enabled by default.
Read →A new analysis examines DeepSeek's agent harness and the broader industry shift toward composable, modular runtimes for building AI agents.
Read →The AI industry, condensed into a five minute read. Free, and you can leave whenever.
Subscribe free