Anthropic Pitches Congress on $1 Billion AI Safety Bill; Meta Debuts Open-Weight Muse Spark
Anthropic seeks to shape federal AI safety law, while Meta releases a powerful open-weight model, intensifying the competitive landscape.
The story
Anthropic has escalated its engagement with Washington, quietly circulating a confidential draft bill to lawmakers that aims to establish federal standards for frontier AI safety and security. This move, reported on October 17, represents a direct attempt by a major AI lab to influence binding legislation, rather than relying solely on voluntary commitments.
The proposed framework, which includes thresholds for investment and personnel, would apply to all frontier labs, including OpenAI, Google DeepMind, Meta, and xAI, effectively setting industry-wide rules. This initiative comes as Anthropic also warned on October 15 that AI models could achieve limited self-replication across internet-connected consumer devices within months, underscoring its focus on safety concerns. The company's proactive legislative push could define the regulatory environment for the entire AI sector.
Who moved
Anthropic
What Changed: Circulated a confidential draft bill on AI safety to Congress and warned of potential AI self-replication across devices.
Consequence: This marks a direct effort to shape federal regulation and highlights its ongoing focus on advanced safety concerns.
Meta
What Changed: Released Muse Spark Open, an 816-billion-parameter open-weight multimodal model on October 15.
Consequence: This aims to bolster its position in the open-weight AI field, particularly for coding and agentic tasks.
NVIDIA
What Changed: Introduced Nemotron 3.5 Super VL and two Bolt edge models, and partnered with Danaher for autonomous AI labs on October 15.
Consequence: These releases expand its open-weight model offerings for agents and deepen its collaboration in scientific research.
Google DeepMind
What Changed: Underwent a reorganization as its Discovery Loop initiative launched on October 15.
Consequence: This internal restructuring suggests a strategic realignment for future AI development.
Kling AI (Kuaishou Technology)
What Changed: Officially launched Kling 4.0, a new video generation model on October 15.
Consequence: This enhances its capabilities in realistic, controlled, and professional-grade AI video creation.
New models
Muse Spark Open
Lab: Meta
What: An 816-billion-parameter open-weight multimodal mixture-of-experts model, accepting text, images, audio, and video input.
Use: Designed for coding and long-horizon agentic work, positioning Meta against Chinese open-weight models.
Nemotron 3.5 Super VL
Lab: NVIDIA
What: A 120-billion-parameter open model offering frontier-level intelligence for agentic workflows, claimed to be 5x faster than competitors.
Use: Aims for fastest task completion and highest accuracy for specialized, always-on enterprise agents.
Nemotron 3.5 Bolt 74B-A9B
Lab: NVIDIA
What: An edge model optimized to balance intelligence and speed for local agentic workloads.
Use: Enables users to delegate projects to local agent harnesses without sending workloads to the cloud.
Kling 4.0
Lab: Kling AI (Kuaishou Technology)
What: A new-generation video generation model with enhanced realism, control over camera movements, and prompt input up to 8,000 tokens.
Use: Provides professional-grade video output for creators and will soon be available to enterprise customers via API.
Market signals
Anthropic's anticipated IPO valuation dropped from $2 trillion to approximately $1.5 trillion on October 15.
Implication: This signals investor caution regarding rising AI financing costs despite the company's Q3 revenue growth.
Iambic Therapeutics raised $150 million in a Nasdaq IPO on October 16, achieving an $800 million valuation.
Implication: This demonstrates continued investor interest in AI-powered drug discovery platforms.
AstraZeneca committed $40 million upfront and up to $1.46 billion in milestones to insitro on October 16.
Implication: This partnership highlights significant pharmaceutical investment in AI platforms for target discovery in complex diseases.
Builders saying
Howard Yu, LEGO Professor of Management and Innovation at IMD Business School. The technological lead of OpenAI and Anthropic over Chinese rivals is 'very perishable' because many use cases do not require the most advanced frontier models.
What we'll be watching
- Mistral AI is expected to release the weights for its Large 4 model by October 27.
- NVIDIA's Nemotron 3.5 Bolt E4B 4GB edge model is scheduled for release on October 23.
- Anthropic's upcoming stock issuance in November, with its re-evaluated valuation, will be closely watched.
- The introductory API pricing for Google's Gemini 4 Argon is set to double on December 1, 2026.
Reporting + analyst voices: grounded via Google Search at publish time.