As President Donald Trump prepares to host China's President Xi Jinping next week, experts and lawmakers are raising alarms about the rapid expansion of artificial intelligence. There is growing concern that AI models are becoming too powerful without adequate safeguards. Experts are calling for increased dialogue between the U.S. and China to slow AI development, but Congress faces hurdles — the House is out of session for seven weeks, making immediate action unlikely. On Wednesday, the House approved AI safety legislation, but it was stalled in the Senate on Thursday.
Senator Heinrich has proposed legislation requiring tech centers to cover all their energy costs, adding another dimension to the policy debate. Former DHS official Paul Rosenzweig examines the legal tools the government already possesses to address threats from frontier AI, analyzing how these powers work and where they might fall short in stopping dangerous AI systems.
Meanwhile, Anthropic revealed that its chatbot Claude is leading more than 26 percent of the company's model research and development efforts. About 30,000 AI agents are conducting research and engineering work at Anthropic at any given time. CEO Dario Amodei recently suggested that AI companies should collaborate to set standards and agree on a slowdown to address superintelligence risks.
NVIDIA released AIPerf, a new benchmarking tool for large language model inference at scale. AIPerf is the successor to GenAI-Perf and uses a multiprocessed system to prevent bottlenecks. It supports 15 or more endpoint types including chat, image generation, and responses, along with various load patterns and public datasets like ShareGPT.
In other developments, Vanderbilt University researchers created a framework to help children learn about AI more deeply, focusing on four concepts: being, knowing, acting, and making. A separate study found that reasoning-mode AI groups agreed nearly unanimously — often on incorrect answers — and that simulated consensus did not track collective accuracy. Geriatrician James Deardorff at the University of California San Francisco warns that AI models for older patients need closer inspection, as doctors frequently trust predictions about sepsis, falls, or death without verifying how well algorithms work for older subgroups.
Key Takeaways
- President Donald Trump is set to host China's President Xi Jinping next week amid growing AI safety concerns
- The House approved AI safety legislation, but it was stalled in the Senate and the House is out of session for seven weeks
- Senator Heinrich proposed legislation requiring tech centers to cover all their energy costs
- Anthropic's Claude leads 26 percent of the company's model research and development efforts
- About 30,000 AI agents are doing research and engineering work at Anthropic at any one time
- NVIDIA released AIPerf, a benchmarking tool supporting 15 or more endpoint types for large language model inference
- A study found reasoning-mode AI groups agreed nearly unanimously, often on incorrect answers
- Former DHS official Paul Rosenzweig examines legal tools available to address frontier AI threats
- Vanderbilt University researchers created a framework with four concepts to help children understand AI
- Geriatrician James Deardorff warns AI models for older patients need closer inspection
AI Safety Concerns Rise as Trump Prepares to Meet Xi Jinping
Experts and lawmakers are warning about the risks of artificial intelligence as President Donald Trump prepares to host China's President Xi Jinping next week. There is concern that AI models are becoming too powerful. Experts want more dialogue between the U.S. and China to slow AI expansion. Congress is urged to set AI safety rules, but the House is out of session for seven weeks, so quick action is unlikely.
Lawmakers Push for AI Safety Rules Ahead of China Meeting
Warnings about artificial intelligence are growing as President Donald Trump prepares to host China's President Xi Jinping next week. There is concern that AI models are becoming too powerful. On Wednesday, the House approved legislation, but it was stalled in the Senate on Thursday. Senator Heinrich proposed legislation requiring tech centers to cover all their energy costs. The House is out of session for seven weeks, making immediate action unlikely.
US and China Face AI Safety Questions Before Trump Xi Summit
Warnings about artificial intelligence are growing as President Donald Trump prepares to host China's President Xi Jinping next week. There is concern that AI models are becoming too powerful. Experts are calling for more dialogue between the U.S. and China to slow the rapid expansion of AI. Next week, Xi is scheduled to meet with Trump, but it is unclear if the two nations can work together effectively. Congress is urged to act on AI safety, but the House is out of session for seven weeks.
Congress Urged to Act on AI Safety as Trump Xi Meeting Nears
Warnings about artificial intelligence are growing as President Donald Trump prepares to host China's President Xi Jinping next week. There is concern that AI models are becoming too powerful. Experts want more dialogue between the U.S. and China to slow AI expansion. On Wednesday, the House approved legislation, but it was stalled in the Senate on Thursday. Senator Heinrich proposed legislation requiring tech centers to cover all their energy costs. The House is out of session for seven weeks, so immediate action appears unlikely.
AI Risks Dominate Discussion as Trump Prepares for China Visit
Warnings about artificial intelligence are growing as President Donald Trump prepares to host China's President Xi Jinping next week, according to an update from September 18, 2026. There is concern that AI models are becoming too powerful. Experts are calling for more dialogue between the U.S. and China to slow the rapid expansion of AI. Congress is urged to establish guardrails for AI safety, but with the House of Representatives out of session for seven weeks, immediate action appears unlikely.
Anthropic uses Claude AI to help build its next model
Anthropic announced that its AI model Claude is helping the company develop the next version of itself. Claude leads 26 percent of Anthropic's model research and development. The company said this means Claude can complete most of a given task. The announcement was made on Thursday.
Anthropic reveals Claude leads AI development work
Anthropic's chatbot Claude is leading more than a quarter of the company's research and development efforts. The company revealed this on Thursday. About 30,000 AI agents are doing research and engineering work at Anthropic at any one time. CEO Dario Amodei recently suggested AI companies should collaborate to set standards and agree on a slowdown to address superintelligence risks.
Teaching kids to question AI not just use it
Vanderbilt University researchers Golnaz Arastoopour Irgens and Sarah K. Burriss created a framework to help kids learn about AI in a deeper way. Their approach includes four concepts: being, knowing, acting and making. The framework encourages students to think about who designs AI and what the consequences are. It also helps children consider what power they have to influence AI systems.
AI groups overstate agreement in reasoning tasks
A study replayed 100 human groups with large language model agent groups on a reasoning task. Agent groups showed more consensus than human groups across multiple measures. About one fifth of human participants never posted during discussions. Reasoning-mode AI groups agreed nearly unanimously, mostly on incorrect answers. The study found simulated consensus did not track collective accuracy.
NVIDIA AIPerf benchmarks AI models at scale
NVIDIA released AIPerf, a new tool for benchmarking large language model inference at scale. AIPerf is the successor to GenAI-Perf and uses a multiprocessed system to prevent bottlenecks. It supports 15 or more endpoint types including chat, image generation and responses. The tool also supports various load patterns and public datasets like ShareGPT.
Paul Rosenzweig examines government power over dangerous AI
Former DHS official Paul Rosenzweig explains the legal tools the government already has to address threats from frontier AI. He discusses how these powers work and where they might fall short. The analysis focuses on whether Washington can effectively stop dangerous AI systems.
Geriatrician calls for careful review of AI for older adults
James Deardorff, a geriatrician at the University of California San Francisco, says AI models for older patients need closer inspection. He notes that doctors often trust predictions about sepsis, falls, or death without checking how well the algorithms work for older subgroups. He developed several prediction models for older adults including mortality and nursing home needs.
AI leaders should act more like parents, opinion argues
A Washington Post opinion piece argues that AI leaders need a course in parenthood. The article responds to recent reports of rogue AI agents and warnings about human extinction. It suggests that AI development could benefit from more responsible and careful oversight from its creators.
Minnesota should shape AI laws with balanced approach
Joel Crandall of the Minnesota Technology Association writes that Minnesota should regulate AI outcomes rather than the technology itself. He notes that Minnesota already has laws on deepfakes in elections, data privacy, and automated decisions in health insurance. He urges candidates to discuss what they hope AI can do for Minnesotans. He says the state is well equipped to handle AI policy through fact-based dialogue.
AI agent harness determines whether security works
In March 2026, LangChain improved its coding agent ranking on Terminal Bench 2.0 by 25 places using only harness engineering. An AI agent harness includes context, tooling, orchestration, verification, and governance layers. The article explains that harness engineering is control system design, not prompting. It also discusses how a poorly engineered harness wastes tokens and why security harnesses are harder than coding harnesses, featuring Cycode as a leader in Agentic Development Security.
Sources
- Concerns growing about AI safety and its capabilities ahead of China-US meeting
- Concerns growing about AI safety and its capabilities ahead of China-US meeting
- Concerns growing about AI safety and its capabilities ahead of China-US meeting
- Concerns growing about AI safety and its capabilities ahead of China-US meeting
- Concerns growing about AI safety and its capabilities ahead of China-US meeting
- Anthropic says its model Claude is helping to build the next version of itself
- Anthropic: AI model Claude helping build next version of it
- AI is shaping kids. What if kids could shape AI, too?
- Language-model groups overstate consensus when replaying human deliberation on a reasoning task
- Benchmarking LLM Inference at Scale with AIPerf
- Does Washington have the power to stop dangerous AI?
- A geriatrician explains why AI for older adults deserves careful scrutiny
- AI leaders need a course in parenthood. It could save us all.
- Opinion | AI is too important to fear, and too powerful to use carelessly
- AI Agent Harness: The Layer That Decides Whether AI Security Works
Comments
Please log in to post a comment.