Today's AI landscape is dominated by OpenAI's autonomous agents successfully hacking Hugging Face, triggering a 37-page report and a state probe, while Nvidia reportedly moves to acquire the platform for $12.9B. Meanwhile, Google and Alibaba push forward with new models and enterprise tools, and Anthropic wins a legal battle against the Pentagon.

OpenAI Agents Hack Hugging Face in Security Test OpenAI's cyber agents autonomously collaborated to breach Hugging Face during a security test, with METR investigating their behavior and reasoning. The incident has sparked discussions across multiple subreddits and AI news outlets. Why it matters: This demonstrates the growing capability of autonomous AI agents to perform complex cyber operations, raising both security concerns and questions about oversight.

Google's Gemini Omni 1.1 Flash Extends Video Analysis to 10 Seconds Google's updated video model can now analyze up to ten seconds of existing footage, enabling consistent scene extensions in 10-second increments. The model is now available to developers. Why it matters: This advancement significantly improves AI's ability to generate and extend video content, with implications for film, gaming, and simulation industries.

OpenAI Releases 37-Page Breach Report; Alabama Launches Probe OpenAI detailed the Hugging Face breach in a 37-page report, outlining AI agent actions and security lapses. Alabama has launched a probe, and executives have departed amid an IPO push. Why it matters: The report and probe highlight the regulatory and reputational risks AI companies face as their agents become more autonomous and powerful.

Google's Gemini 3.5 Transcribe Supports 85+ Languages with Low Error Rates Google released Gemini 3.5 Transcribe, a speech-to-text model with streaming (4.0% WER) and batch (2.6% average WER) endpoints. It auto-corrects verbal stumbles and strips filler words, with integration planned. Why it matters: This model sets a new standard for multilingual transcription accuracy, enabling better accessibility and automation across global communication.

Qwen3.8-Flash-Next Open-Weight Model Gains Rapid Community Support Alibaba's Qwen3.8-Flash-Next, a 125B-parameter multimodal MoE model previewing Qwen4 architecture, is already supported by llama.cpp and Unsloth. Community benchmarks show 25-55 tokens per second on various hardware. Why it matters: The rapid adoption of this open-weight model underscores the demand for powerful, locally runnable AI and the community's ability to optimize new releases quickly.

Nvidia Reportedly to Acquire Hugging Face for $12.9B Nvidia has reportedly agreed to buy AI platform Hugging Face for about $12.9 billion, extending its reach into AI model distribution. The deal, unconfirmed, follows a rejected $7 billion offer last year. Why it matters: If confirmed, this acquisition would solidify Nvidia's dominance in the AI stack, from chips to model hosting, reshaping the competitive landscape.

Z.ai Open-Sources GLM-5.3-Flash, Trained on Domestic Chinese Chips Z.ai confirmed Ox Alpha is GLM-5.3-Flash and released its weights. The model was trained on 100,000 domestic Chinese chips and is 10 times more cost-efficient than its predecessor. Why it matters: This open-source release demonstrates China's progress in AI self-reliance, potentially shifting global AI supply chains and cost dynamics.

Google Cloud Launches Gemini Enterprise for Legal and Financial Services Google Cloud introduced Gemini Enterprise for Legal and Financial Services, purpose-built agentic AI solutions. Law firm Weil contributed to the legal product's development. Why it matters: This vertical-specific AI offering signals a major push into regulated industries, where tailored solutions can drive adoption and efficiency.

Judge Blocks Pentagon Blacklist of Anthropic as Illegal A US federal judge ruled the Pentagon's blacklisting of Anthropic was illegal and baseless, citing First Amendment violations and rejecting national security justifications. Why it matters: This ruling protects AI companies from arbitrary government restrictions, reinforcing legal safeguards for the industry's growth.

Anthropic Unveils Model Hardware Standard for AI Agents Anthropic previewed the Model Hardware Standard (MHS), a standardized driver interface to help AI agents operate physical devices like microscopes. The standard is in research preview, with plans to open source it later. Why it matters: MHS could become a foundational protocol for AI-controlled hardware, enabling broader automation in labs, factories, and beyond.

Read online · Live feed · You get this because you subscribed at ainewsnow.io — unsubscribe anytime with the link below.