TODAY'S TOP STORIES - JUNE 5, 2026
- Great American AI Act - Bipartisan 269-page bill would preempt all state AI laws for 3 years. Colorado AI Act (due June 30) directly in its crosshairs. Discussion draft only - not yet law
- Nemotron 3 Ultra on SageMaker JumpStart - Day-zero availability, one-click deploy, NVFP4 for 5x faster inference and 30% lower cost on AWS
- OpenAI Realtime Audio GA - GPT-Realtime-2 (GPT-5-class continuous audio), GPT-Realtime-Translate (70+ languages), GPT-Realtime-Whisper (live streaming transcription). Realtime API exits beta
1. The Great American AI Act Would Block All State AI Laws for 3 Years
Reps. Jay Obernolte (R-CA) and Lori Trahan (D-MA) unveiled the Great American Artificial Intelligence Act on June 4 - a 269-page bipartisan discussion draft built on four pillars: frontier model safety (codifying CAISI), workforce impact tracking, cybersecurity, and AI R&D. The bill was joined by four co-sponsors including Reps. Subramanyam, Franklin, Peters, and Houchin. It comes days after President Trump signed an executive order establishing a National Policy Framework for AI.
The most contentious provision: a three-year preemption of state AI laws. Colorado's AI Act is due to take effect on June 30 - 25 days away. California's growing portfolio of AI laws would also be suspended. More than 200 state lawmakers have already written to Congress opposing preemption, and the provision previously died in two other legislative vehicles (the One Big Beautiful Bill and the NDAA) after a 99-1 Senate vote stripped it out. This is a discussion draft - not yet formally introduced. The bill faces significant opposition from civil rights groups, state governments, and some Republican senators who previously blocked preemption.
Full article: What's in the Great American AI Act, who opposes the preemption, and what happens next ->
2. Nemotron 3 Ultra Is on SageMaker JumpStart with One-Click Deployment
AWS confirmed day-zero availability of Nemotron 3 Ultra on SageMaker JumpStart on June 4 - the same day the model shipped on Hugging Face and NVIDIA NIM. SageMaker JumpStart provides one-click deployment with NVFP4 precision optimization: 5x faster inference throughput and 30% lower cost compared to BF16, thanks to NVIDIA's 4-bit floating-point format optimized for Blackwell GPU tensor cores. The model's hybrid Transformer-Mamba MoE architecture - combining Mamba-2 state-space layers for constant-memory long-context handling with sparse MoE routing - is specifically designed for long-running agentic workloads where an agent plans, executes, and iterates across hundreds of turns at 1M token context lengths.
Full guide: How to deploy Nemotron 3 Ultra on SageMaker, NVFP4 explained, and architecture deep dive ->
3. OpenAI's Realtime Audio API Is Out of Beta - Three New Models GA
OpenAI launched three new real-time audio models in May 2026, and the Realtime API is now generally available for production use. GPT-Realtime-2 is the flagship: built on GPT-5-class reasoning, it processes audio in a continuous stream - no separate transcription step, no text intermediary, no synthesis gap. It can hear tone and interruptions, reason about them, and respond in audio at lower latency than any pipeline-based voice system. GPT-Realtime-Translate adds live multilingual voice translation for 70+ input languages and 13 output languages; Deutsche Telekom and Vimeo are confirmed testers. GPT-Realtime-Whisper adds streaming live transcription, producing captions as speech occurs rather than after recording ends.
Pricing: GPT-Realtime-2 and Translate at $40/$80 per million audio tokens (input/output). GPT-Realtime-Whisper at $3/M input tokens - the accessible entry point for teams that need live transcription without full conversational AI.
Full breakdown: All three models compared, pricing, WebSocket vs WebRTC, and when to use each ->