China's AI models have Trump's AI world at war with itself

Title: US-China AI Rivalry, Chip Wars, and Safety Lessons

01Trump's Advisors Wage War on US AI Companies Over China

The weekend brought an unusual public spectacle as several current and former advisors to President Donald Trump on AI publicly attacked the country's leading AI companies. David Sacks, the president's former AI and crypto "czar," was among those who lobbed insults at American firms, highlighting a deepening rift over how the US should respond to China's AI competition. The conflict centers on disagreements about open-weight model policies, with advisors pushing for a harder line against Chinese AI development while companies resist government mandates.

The dispute reveals a fundamental tension between national security priorities and commercial interests in the AI sector. Tech giants like Google, OpenAI, and Anthropic are caught between pressure to compete globally and regulatory expectations around open-weight model restrictions. China's release of models like Kimi K3 has only intensified these debates, with some advisors calling for bans on Chinese open-weight LLMs while companies warn such measures could backfire.

02Google's New AI Chip Signals Escalating Hardware Race

Alphabet's Google is reportedly developing a new chip designed to make its Gemini models run much more efficiently, signaling escalating hardware competition among major AI players. The custom AI chip aims to optimize model performance while reducing operational costs, positioning Google to better compete with rivals who have invested heavily in proprietary silicon. This development reflects a broader industry trend where companies seek to control more of their AI infrastructure stack.

The move comes as AI labs face mounting pressure to demonstrate sustainable business models amid intense competition. Custom chips like this could provide significant efficiency gains for running large language models, potentially reshaping the economics of AI development. Google joins competitors in the race to develop specialized hardware that can handle the increasing computational demands of frontier AI systems.

03OpenAI Publishes Candid Safety Lessons from Long-Horizon Deployments

OpenAI has published research sharing lessons learned from deploying long-running AI models, marking a notable shift toward transparency about observed failures and new safety risks. The company's safety team documented challenges that emerge when AI systems operate over extended periods, including alignment issues that only surface during prolonged interaction. This research represents one of the most candid admissions from a major AI lab about the practical difficulties of maintaining safe AI systems in real-world deployments.

The findings highlight how iterative deployment can reveal risks that controlled testing misses. OpenAI emphasized that improved safeguards have been developed through this process, demonstrating that transparency and iteration are essential to AI safety. The company acknowledged that long-horizon models present unique challenges that require ongoing vigilance and adaptive safety measures rather than one-time solutions.

Also today

04Trump's latest AI czar has already resigned — The Center for AI Standards and Innovation director role continues to be unstable as leadership turnover persists.
05OpenAI is scared of open-weight models. Should the US be? — Debate intensifies over banning Chinese open-weight LLMs as companies grapple with turning AI into sustainable business models.
06Kimi K3: The open-weights escalation — China's release of Kimi K3 open-weight model escalates global competition and raises questions about US AI strategy.
07AI's most important protocol is getting a little bit easier to use — A stateless approach to session management makes AI protocols more accessible to developers.
08AI is more likely than humans to form biases when hiring — Research reveals LLMs develop independent biases beyond training data, raising concerns about fairness in automated recruitment.
09Human mathematicians are being outcounterexampled — AI systems increasingly generate counterexamples faster than human mathematicians can verify them.
10Introducing Cosmos 3 Edge — NVIDIA releases next-generation edge AI capabilities for robotics and autonomous systems.