Anthropic Proposes New AI Regulation Framework to Congress

Key Takeaways

• Anthropic proposed AI regulation requiring at least 50 deaths or one billion dollars in damages before the government can block a model, with independent evaluators and a D.C. judge making the call
• Anthropic's leaked draft lets AI labs write their own standards, differing from CEO Dario Amodei's earlier FAA-based proposal and drawing criticism from safety advocates
• TrendAI integrated NVIDIA Nemotron models into TrendAI Vision One for Sovereign and Private Cloud, keeping AI processing inside customer-controlled infrastructure
• BBVA partnered with Nvidia to embed security into AI agents from the start, aiming to set a new standard for AI security in finance
• The UK relaunched its semiconductor center as the Semiconductor Catapult with labs in Newport, Bristol, and Glasgow, backed by a £1.1bn AI Hardware Plan
• A machine learning method called learned control variates could cut city-wide blackout days from EV charging grid overloads by a third, based on tests in Surabaya, Indonesia
• The TURTLE benchmark uses three LLM agents to evolve AI tool-use tasks without human help, producing harder and more diverse test suites
• The University of Michigan released six principles for responsible AI use, stressing that AI should support rather than replace human skills
• The global AI market could reach $150 billion by 2030 with 30 percent annual growth, but a talent gap of 4.7 million is expected by 2030
• Maryland Governor Wes Moore highlighted $25 million in new funding tied to H.R. 1 at the Bloomberg American Health Summit, while Baltimore's murder rate fell 60 percent in three years

Anthropic proposes AI regulation with 50-person death threshold

Anthropic proposed a bill to Congress that would set rules for AI. The government could only block an AI model if it posed an imminent risk. That risk must involve at least 50 deaths or one billion dollars in damages. The danger must link to biological weapons, model escapes, or cyberattacks. Independent evaluators under the Commerce Department would decide if conditions warrant halting a model.

Anthropic's leaked AI bill differs from FAA model

Anthropic's leaked draft AI legislation differs from the Federal Aviation Administration model that CEO Dario Amodei proposed earlier. Under this framework, AI labs would write the standards they must follow. Some AI safety advocates want the government to set those standards instead. The leak also suggests Anthropic may be trying to gain favor with the Trump administration on AI policy.

Anthropic AI safety bill could block models threatening 50 or more lives

Anthropic proposed a framework for government regulation of AI. A model could be blocked if it posed a foreseeable risk killing over 50 people or causing one billion dollars in damage. The risk must stem from biological weapons, model escapes, or cyberattacks. Independent evaluators registered with the Commerce Department would assess the risk. A D.C. judge would need to approve any halt on a model's release.

TrendAI brings AI security to sovereign clouds with NVIDIA Nemotron

TrendAI announced it is integrating NVIDIA Nemotron open models into TrendAI Vision One for Sovereign and Private Cloud. The integration brings AI-assisted security operations into customer-controlled environments. It supports guided investigation, security reasoning, and agentic SIEM. The product is available today for private cloud, sovereign cloud, and on-premises deployment. Customers maintain control over where data and AI processing reside.

TrendAI runs NVIDIA Nemotron inside customer-controlled infrastructure

TrendAI launched TrendAI Vision One for Sovereign and Private Cloud, running NVIDIA Nemotron models entirely inside customer infrastructure. This delivers AI-driven threat detection without security data leaving the sovereign boundary. TITECH is working with TrendAI as a design partner for customers in Oman. Bechtle AG praised the integration for supporting secure AI adoption at scale. The Nemotron integration supports policy-based response and guided security operations.

UK launches Semiconductor Catapult for AI and defense tech

The UK relaunched its Compound Semiconductor Applications Catapult as the Semiconductor Catapult with a broader mission. It will support AI hardware, defense, and other key sectors through labs in Newport, Bristol, and Glasgow. The move follows the government's £1.1bn AI Hardware Plan. The center aims to help companies test and validate semiconductor designs. It also seeks to keep more economic value from semiconductor innovation within the UK.

AI model helps estimate rare power grid outages at EV charging stations

Electric vehicle charging can add heavy load to power grids, and rare overloads can cause city-wide blackouts. Researchers developed a machine learning method called learned control variates to better estimate these rare outage risks. The approach uses cheap features like temperature and load forecasts to reduce the number of costly simulations needed. Tests in Surabaya, Indonesia showed the method could cut city-wide blackout days by a third. Even a weak learner removed about 25 percent of uncertainty, while a stronger model removed about 60 percent.

TURTLE benchmark evolves AI tool-use tasks without human help

AI assistants need stronger tool-use skills to work well with apps like Gmail, Venmo, and Slack. Researchers introduced TURTLE, an unsupervised method that evolves existing tasks into harder and more diverse test suites. It uses three large language model agents: a proposer, a solver, and a feedback generator. Tests on AppWorld, tau-Bench, and AutomationBench with Gemma 4 31B-it showed more hard tasks and broader failure modes. The approach aims to make tool-use benchmarks easier to scale across different environments.

Public health forum addresses guns, drugs, AI and budget cuts

The 2026 Bloomberg American Health Summit in Baltimore brought together over 500 public health leaders. Topics ranged from traditional issues like drug overdoses to newer concerns such as gun violence, AI, social media, and federal funding cuts. Maryland Governor Wes Moore and Baltimore Mayor Brandon Scott spoke about protecting healthcare access. Moore highlighted $25 million in new funding tied to H.R. 1 and a new Vax Act. Scott noted Baltimore's murder rate fell 60 percent in three years but said more work remains.

AI buildout is 15 percent done, report says

A MarketBeat article says the global AI buildout is still in early stages and points to future investment areas. A McKinsey report estimates the global AI market could reach $150 billion by 2030 with about 30 percent annual growth. Growth is expected mainly from AI use in healthcare, finance, and manufacturing. However, the report warns of a global AI talent gap reaching 4.7 million by 2030. Future spending is likely to focus on new AI technologies and hiring skilled workers.

BBVA partners with Nvidia to make AI agents safer

BBVA is working with Nvidia to build security into AI agents from the very beginning. AI agents are software programs that work on their own, but they can be hit by cyber attacks. The two companies want to create safer AI tools that can spot and block threats before they cause harm. This effort aims to set a new standard for AI security in the finance industry. The partnership reflects BBVA's goal to stay ahead in technology innovation.

University of Michigan sets 6 rules for AI use

The University of Michigan released six principles for using artificial intelligence in a responsible way. The ideas came from a faculty survey, research and feedback from students and staff. The rules say AI should support human skills rather than replace them. They also call for strong ethics, transparency, privacy protection and ongoing improvement. The university wants students to keep learning through real challenges while building their AI knowledge.

THMY grows as demand rises for AI testing

THMY Holdings has seen its share price jump nearly 7.5 times in under a year, reaching RM2.32. The company provides automated test solutions for electronic products and is now involved earlier in the design of AI chips and data centre equipment. With about 200 employees, THMY helps customers test how products handle heat and power before they go into production. The company is also eyeing physical AI and humanoid robots as future growth areas. THMY is planning to expand its factory using bank loans and possibly equity funding.

Sources

• Anthropic’s Proposed Regulations for AI Include a Death Threshold of at Least 50 People Before Halting Models
• AI & Tech Brief: Anthropic’s leaked bill
• Anthropic pitches AI safety bill that may block a model if 50+ people could be killed
• TrendAI™ Brings AI-Driven Security That Runs Entirely Under Customer Control to Sovereign and Private Clouds with NVIDIA Nemotron™ Open Models
• TrendAI™ Brings AI-Driven Security That Runs Entirely Under Customer Control to Sovereign and Private Clouds with NVIDIA Nemotron™ Open Models
• UK launches expanded Semiconductor Catapult targeting AI hardware and defense
• Learned Control Variates for Rare Grid-Outage Estimation at Electric Vehicle Charging Stations
• Evolving Self-tuning Benchmarks for Long-Horizon Tool Use
• Guns, drugs, AI, budget cuts join traditional maladies at public health forum
• The AI Buildout Is Only 15% Complete—Here Is Where the Next Dollar Actually Goes
• BBVA works with Nvidia to boost AI agent security
• University of Michigan Outlines 6 Guidelines for Using AI
• THMY scales up for next generation of AI testing