Daily AI News - 2026-06-06

⏰ Limited-Time Free Models

Several limited-time free and discounted models are nearing expiration:

The DeepSeek V3.1 Nex N1 deal expires in just 1 day — grab it while available.

🚨 Anthropic Calls for Global AI Pause — Warns of “Recursive Self-Improvement”

Anthropic triggered a firestorm of debate today by publicly calling for a pause on global AI development, warning that frontier models are approaching a dangerous milestone: recursive self-improvement, where AI systems enhance their own capabilities without human intervention.

In a coordinated blitz of media coverage — including the New York Times, CNN, Fortune, WSJ, Scientific American, and Tom’s Hardware — Anthropic urged governments and AI labs to establish a “pause button” that would halt frontier development before humans risk “losing control.” The company’s blog post warned that Claude’s internal development speed has been accelerating faster than anticipated, and that self-improving AI loops could emerge within current training paradigms.

Meanwhile, Reuters reported that tensions between Anthropic and the White House have eased ahead of the company’s anticipated IPO, with both sides engaging in behind-the-scenes negotiations. The IPO filing (S-1) has reportedly influenced the timing of Anthropic’s unusually public alarm bells.

🏛️ Trump Signs Scaled-Back AI Executive Order — Voluntary 30-Day Review

President Trump signed a scaled-back AI executive order today that establishes a voluntary federal vetting program for frontier AI models. Under the order, AI developers can submit new models for a 30-day national security review before public release.

Key provisions:

• Voluntary, not mandatory — a departure from earlier drafts that proposed mandatory government access
• Focused on national security risks from frontier AI systems
• IBM CEO backed the order, calling it “balanced and pragmatic”
• White House will meet with AI companies on a government profit-sharing plan as soon as next week

The order represents a middle ground between industry self-regulation and the mandatory AI licensing regime some lawmakers had sought. Critics argue the voluntary framework lacks teeth, while supporters say it establishes essential protocols without stifling innovation.

🤝 OpenAI Diverges from White House, Offers Government Early Access

In a move that contrasts with Anthropic’s adversarial stance, OpenAI has offered the US government early access to frontier models before public release. The proposal comes as Politico reports OpenAI is diverging from the White House on AI safety rules, preferring a cooperative “insider” approach rather than external regulation.

The company also revealed that AI systems are now being used to design OpenAI’s next-generation models, a development SoftBank CEO Masayoshi Son called a sign of “superintelligence” in a CNBC interview.

💰 Alphabet Raises $85B for AI Build-Out — Berkshire Invests $10B

Alphabet announced plans to raise $80–85 billion through stock sales to fund its AI infrastructure build-out, with Berkshire Hathaway committing $10 billion of that total. This is the largest single AI capex raise by any company to date.

Goldman Sachs estimates that AI spending from the big four tech giants (Meta, Microsoft, Amazon, Alphabet) will exceed Japan’s GDP through 2030. Meta, Microsoft, Amazon, and Alphabet are spending at a pace that Forbes called “shocking” relative to current AI revenue generation.

🚀 DeepSeek Previews V4, Makes 75% Price Cut Permanent

China’s DeepSeek is on a roll:

• DeepSeek V4 preview released — TechCrunch reports the model “closes the gap with frontier models”
• 75% price cut on V4-Pro made permanent — significantly undercutting Western competitors
• $7 billion maiden funding round — reportedly led by Tencent, signaling massive investor appetite

The V4 preview comes just weeks after DeepSeek’s V3.1 series and positions the company as a serious contender in the global AI race, especially on price-performance.

🧠 Moonshot AI Releases Kimi-K2.6 — 1T Parameters, IPO on the Horizon

Moonshot AI released Kimi-K2.6, a 1-trillion-parameter model with attention optimizations that push the frontier of long-context performance. The model follows the K2.5 release and shows continued refinement of Moonshot’s open-weight strategy.

In parallel, the company is reportedly considering a Hong Kong IPO after raising $2 billion at a $20 billion valuation. Bloomberg reports that Moonshot is moving to unwind its offshore structure — a prerequisite for listing in Hong Kong.

🏎️ Google to Pay SpaceX $920M/Month for Compute

In a stunning exclusive by TechCrunch, Google is reportedly paying SpaceX $920 million per month for compute capacity — likely leveraging SpaceX’s Starlink-related data center infrastructure and satellite capabilities for AI workloads. The deal, if confirmed, would be one of the largest single compute procurement agreements in history.

In Brief

• Meta delays Muse Spark API release — The New York Times reports Meta is delaying its new AI model rollout over performance concerns, just weeks after hiring Scale AI’s Alexandr Wang.
• Florida sues OpenAI and Sam Altman — Politico reports Florida has filed a lawsuit against OpenAI and Sam Altman over AI risks, adding to the growing regulatory pressure.
• Connecticut enacts comprehensive AI regulation — Becomes the latest state to pass broad AI legislation, joining Illinois and Colorado.
• House lawmakers propose federal bill to preempt state AI rules — A draft would prohibit states from enacting their own AI laws for three years.
• Nvidia unveils new PC chips — Jensen Huang’s GTC Taipei keynote positions Nvidia to dominate every layer of the AI stack from cloud to edge.
• Apple approves Poke as first AI agent on Messages for Business — A notable step toward conversational commerce AI.
• Grok Imagine 1.5 Preview — xAI released a preview of its image generation model with new capabilities.
• AI-designed universal coronavirus vaccine completes first human trial — A breakthrough in AI-driven drug discovery.
• AirTrunk commits $30B for 5GW of AI data centers in India — Massive infrastructure bet on India as an AI hub.
• NSA publishes MCP security guidance — Design considerations for securing Model Context Protocol deployments.

🤗 New Hugging Face Models

Notable uploads on the HF Hub:

• donate110/evolai_test — Qwen3-based experimental model, 4,310 downloads.
• juergengunz/fluxer — Flux-related model with 8 likes, gaining community attention.
• elicohen26/vulnai-mistral-7b-merged — Security-focused Mistral fine-tune for vulnerability AI.
• mradermacher/granite-4.1-8b-apostate-i1-GGUF — GGUF quantized Granite variant.
• minimew/text2sql-checkpoints — Text-to-SQL model checkpoints.

🚀 New Models on OpenRouter

Four additions:

• NVIDIA: Nemotron 3.5 Content Safety (free) — $0/M tokens, 128K context. Dedicated content safety model.
• NVIDIA: Nemotron 3 Ultra (free) — $0/M tokens, 1M context. Massive 550B-55B MoE architecture.
• NVIDIA: Nemotron 3 Ultra — $0.5/M tokens input, 1M context.
• Qwen: Qwen3.7 Plus — $0.4/M tokens input, 1M context.

⭐ GitHub Trending: AI Edition

💡 Key Trends

1. The Safety vs. Speed Debate Escalates — Anthropic’s global pause call, Trump’s voluntary review order, state-level regulation, and OpenAI’s divergent approach all point to an increasingly fragmented governance landscape.

2. Infrastructure Spending Reaches New Heights — Alphabet’s $85B raise, Google’s $920M/month SpaceX deal, AirTrunk’s $30B India bet, and $70B ByteDance capex signal that AI infrastructure investment has entered a new, unprecedented phase.

3. Asian AI Labs Compete on All Fronts — DeepSeek (China) with V4 preview and 75% price cuts, Moonshot (HK) with 1T-parameter Kimi-K2.6, Alibaba’s Qwen expanding into automotive and enterprise agents — Asian AI companies are no longer following, they’re leading in key segments.