Daily AI News - 2026-06-08

🧠 Anthropic Urges Global Pause in AI Development Over ‘Recursive Self-Improvement’ Risks

Anthropic set off alarm bells across the industry on Monday, issuing a formal call for a global pause on frontier AI development. In statements covered by The Wall Street Journal, Reuters, and The Guardian, the company warned that its Claude AI systems are showing signs of recursive self-improvement — where AI models autonomously enhance their own capabilities faster than their creators can measure or control.

“Humans risk losing control of AI if this trajectory continues,” Anthropic stated, urging all major AI labs to pause development. SoftBank CEO Masayoshi Son countered in an interview with CNBC that AI is already designing OpenAI’s next model, calling it a sign of approaching “superintelligence.”

Meanwhile, over 150 mathematicians from top universities published an open letter warning governments not to “believe the hype” about AI’s mathematical reasoning capabilities, arguing that benchmarks are routinely gamed and that LLMs fundamentally lack true logical understanding.

⚡ Florida Sues OpenAI and Sam Altman, Seeks Personal Liability

The State of Florida filed a sweeping lawsuit against OpenAI and CEO Sam Altman, alleging the company knowingly concealed ChatGPT’s safety risks from users. The suit, covered by Politico, NPR, and CNBC, seeks to hold Altman personally liable for alleged harms, citing “utter disregard for the risk to human life.”

Florida Attorney General argues the company misled the public about the model’s capabilities and risks. This is the first major state-level lawsuit targeting an AI CEO with personal liability claims, marking a significant escalation in AI-related litigation.

🏛️ Trump Signs AI Executive Order, AI Adviser Departs, Government Bans Anthropic

President Trump signed a new executive order establishing a voluntary federal vetting program for frontier AI models, requiring companies to submit models for national security review before public release. The order, reported by The Guardian and AP News, allows the federal government early access to new AI releases.

In a separate development, White House AI policy adviser Sriram Krishnan — a key figure in the administration’s AI strategy — is leaving his role, per The Washington Post and TechCrunch.

In a dramatic escalation of the Anthropic-Pentagon dispute, the Trump administration ordered all federal agencies and military contractors to cease business with Anthropic, as reported by CNN and the BBC. The order follows Anthropic’s refusal to remove safety guardrails from its models for Pentagon use. OpenAI promptly announced a new Pentagon deal to fill the gap.

💰 DeepSeek Targets $7.4B Raise at $59B Valuation; Moonshot AI Seeks $30B

Chinese AI labs are raising enormous war chests. DeepSeek is slated to raise $7.4 billion in its first-ever funding round at a valuation of up to $59 billion, with Tencent and CATL among the backers (Reuters). The company’s low-cost AI models are also topping US enterprise spending indexes, with Ramp reporting DeepSeek as the fastest-growing software vendor among US businesses in June 2026.

Meanwhile, Moonshot AI (maker of the Kimi chatbot) is seeking a $2 billion fundraising round at a valuation of up to $30 billion, per Bloomberg, reflecting surging global demand for open-source and low-cost AI alternatives.

🤖 Microsoft Unveils 7 In-House AI Models at Build 2026

At Microsoft Build 2026, the company unveiled a new family of seven in-house MAI (Microsoft AI) models spanning reasoning, voice, coding, and image generation — a major strategic step to reduce reliance on OpenAI and lower costs for developers (CNBC, Windows Central). The models cover the full spectrum from small efficient agents to large frontier-capability systems, and are available immediately on Azure.

🍎 Apple WWDC 2026: Gemini-Powered Siri Debuts

Apple’s WWDC 2026 keynote featured the launch of a Google Gemini-powered Siri overhaul — the most significant update to the voice assistant since its introduction. According to MacRumors and The Verge, the new Siri leverages Gemini’s advanced language understanding for more natural conversations, contextual awareness, and third-party app integration. Tim Cook’s final keynote as CEO also introduced iOS 27 and macOS 27.

📋 In Brief

• Meta is delaying the Muse Spark AI model API release due to bugs and infrastructure issues (WSJ).
• NVIDIA and SK hynix announced a multiyear technology partnership to develop advanced memory for AI factories (Reuters).
• OpenAI is planning a major ChatGPT “superapp” overhaul featuring coding tools and AI agents ahead of its listing (Financial Times).
• Anthropic expanded Project Glasswing / Claude Mythos to 150 additional organizations across 15+ countries, including critical infrastructure operators (CNBC).
• xAI launched Grok Build 0.1 — its first AI coding agent — in early beta for subscribers, with a new model tripling parameter count expected mid-June (PCMag).
• A bipartisan House draft bill proposes a three-year preemption of state AI laws, sparking debate over federal vs. state AI regulation (Roll Call, Politico).
• SoftBank CEO Masayoshi Son stated that AI is already designing OpenAI’s next model, calling it a milestone toward artificial superintelligence (CNBC).

⏰ Limited-Time Free Models

Expiring in 2 days: DeepSeek R1 Distill Llama 70B and NVIDIA Nemotron Nano 9B V2.

🤗 Trending on Hugging Face

• OpenTransformer/AGILLM-4 — A long-context language model with 8 likes, gaining attention for its AGI-LLM architecture.
• ViteMamba/TigerLLM-Medical-Bengali — A Gemma3-based medical LLM fine-tuned for Bengali, showcasing growing interest in regional LLMs.
• Plus 8 region-tagged experimental uploads — HuggingFace activity was moderate today with several zero-download experimental models.

⭐ GitHub Trending: AI Edition

• jd-opensource/JoyAI-Echo ⭐876 — Pushing the frontier of long audio-visual generation with synchronized audio and extended video outputs.
• tastyeffectco/sandboxd ⭐499 — Self-hosted dev sandboxes with preview URLs, built in Go for coding agents and SaaS factories.
• JimLiu/baoyu-design ⭐346 — Run Claude Design locally as an Agent Skill; produces polished UI mockups as self-contained HTML.
• amElnagdy/guard-skills ⭐343 — Quality gates that catch AI-generated failure modes in code, tests, and documentation.
• razr001/align-dev ⭐274 — Helps AI-assisted frontend teams generate shared coding standards and SKILL.md files for consistent agent output.