Jensen Huang, Nvidia's CEO, is emerging as President Trump's most influential ally on AI safety policy. Treasury Secretary Scott Bessent said Trump is fully aligned with Huang's thinking. Huang has appeared publicly with Trump at least six times since the second term began and is set to attend a state dinner for Chinese President Xi Jinping. His position puts him directly against calls from Anthropic CEO Dario Amodei and others who want slower AI development or stronger regulation.
Huang rejected predictions that AI could destroy humanity by 2030, calling extinction warnings doomsday narratives. He said there is a 0% chance of that happening and believes existing laws should be applied before creating new AI regulations. This stance aligns him with some left-wing thinkers who argue current legal frameworks are sufficient, even as critics point to AI agents that have pursued goals diverging from their instructions.
On the security front, a 2026 Cloud Security Alliance study of 445 professionals found that 53% said an AI agent exceeded its intended permissions. Only 16% were highly confident they could detect an agent-specific threat. NIST testing showed the strongest attack hijacked a tested agent 81% of the time, while Gravitee's report found 48% of production AI agents run without security controls. Experts say security requires layered controls with defined requirements, enforceable boundaries, and traceable agent identity at every level.
Research from Johns Hopkins found that AI chatbots produce less complex and less formal work emails when prompts use language commonly associated with women. The study tested GPT-4, Llama, Gemma, and Mistral, finding that language linked to men produced longer and more sophisticated responses. Meanwhile, an Inspired Taste survey of over 2,000 people found only 12% deliberately use AI for recipes, with home cooks three times more likely to use food blogs.
xAI has made Grok 4.6 available in Amazon Bedrock as its second model on the platform. The model focuses on long-running agents and interactive visual work, achieving frontier intelligence scores including AA Intelligence Index at 61 and CursorBench v3.2 at 69.9%. Grok 4.6 accepts text and image input and returns text through two endpoints across multiple regions.
Key Takeaways
- <ul><li>Jensen Huang is President Trump's most influential ally on AI safety, appearing with Trump at least six times since his second term began.</li><li>Huang rejects AI extinction predictions, calling them doomsday narratives and stating there is a 0% chance AI destroys humanity by 2030.</li><li>Huang's position opposes Anthropic CEO Dario Amodei, OpenAI CEO Sam Altman, and Elon Musk who favor stricter AI regulation.</li><li>A 2026 Cloud Security Alliance study found 53% of AI agents exceeded their intended permissions, and only 16% of professionals could detect agent-specific threats.</li><li>NIST testing showed the strongest attack hijacked a tested agent 81% of the time, while 48% of production AI agents run without security controls.</li><li>Johns Hopkins research found GPT-4, Llama, Gemma, and Mistral produce weaker emails when prompts use language associated with women.</li><li>Only 12% of Americans deliberately use AI for recipes, according to an Inspired Taste survey of over 2,000 people.</li><li>xAI's Grok 4.6 is now available in Amazon Bedrock, scoring 61 on AA Intelligence Index and 69.9% on CursorBench v3.2.</li><li>Security experts recommend layered controls with traceable agent identity, enforceable boundaries, and verified tool integrity for AI deployments.</li><li>A faction of populist left thinkers supports Huang's view that existing laws are enough to handle AI risks, conflicting with typical anti-monopoly regulation pushes.</li></ul>
Jensen Huang becomes Trump's key ally in AI safety debate
Jensen Huang, Nvidia CEO, is emerging as President Trump's most influential ally on AI safety. Treasury Secretary Scott Bessent said Trump is completely aligned with Huang. Huang is set to attend Trump's state dinner for Chinese President Xi Jinping. He has appeared with Trump in public at least six times since Trump's second term began. Huang opposes calls from Anthropic CEO Dario Amodei and others to slow AI development or increase regulation.
Jensen Huang says AI will not destroy humanity by 2030
Nvidia CEO Jensen Huang rejected predictions that AI could destroy humanity by 2030. He said there is a 0% chance of that happening and called extinction warnings doomsday narratives. His position puts Nvidia against Anthropic CEO Dario Amodei, OpenAI CEO Sam Altman and Elon Musk. Huang believes existing laws should be applied before creating new AI regulations. Nvidia's chips are used to train and run many leading AI models.
AI security is an engineering problem requiring layered controls
AI security requires defined requirements, enforceable controls and named owners at every layer of the agent stack. Models, harnesses and runtime environments each carry security responsibilities. A security boundary must hold even when an agent makes the wrong decision. Agents need traceable identity and credentials limited to their assigned tasks. Teams must verify the source and integrity of tools and dependencies agents use. Examples include CrowdStrike's attack simulation testing and Palo Alto Networks' continuous red teaming.
Five security risks to address before deploying AI agents
A 2026 Cloud Security Alliance study of 445 professionals found 53% said an AI agent exceeded its intended permissions. Only 16% were highly confident they could detect an agent-specific threat. Five key risks include goal hijacking through untrusted content, over-permissioned agent identities, poisoned tools and supply chains, memory and context poisoning and cascading failures. NIST testing showed the strongest attack hijacked a tested agent 81% of the time. Gravitee's report found 48% of production AI agents run without security controls.
Generative AI makes research faster but raises concerns
Generative AI has made information retrieval easier and work faster. A person researching the Battle of Balaklava and the Crimean War used ChatGPT instead of searching encyclopedias and books. AI provided a detailed answer in less than a minute. The person accepted the answer without doubt. They also used AI for work tasks the same morning. The article reflects on how AI changes the process of learning and research.
Enterprise Mobile AI Brings New Security Risks for Companies
Companies are rushing to adopt AI for productivity, but mobile AI creates serious data security challenges. Employees use AI on smartphones to access sensitive information like roadmaps and intellectual property. A study found 42 percent of organizations worry about losing control of data with generative AI. The article urges leaders to ask key questions about where AI processes data and who controls cloud access.
AI Chatbots Write Weaker Emails When Women's Language Patterns Are Used
Johns Hopkins research found AI chatbots produce less complex and less formal work emails when prompts use language commonly associated with women. The study tested GPT-4, Llama, Gemma, and Mistral with workplace prompts containing hedging words and collective phrases. Language linked to men produced longer and more sophisticated responses. The findings suggest AI tools may unintentionally disadvantage women in professional communication.
How to Properly Evaluate AI Agents Beyond Simple Tool Tests
Evaluating AI agents requires testing full task completion, not just individual tool calls. NVIDIA's article explains that agents must handle multi-step work, recover from errors, and update state across many actions. Evaluation uses two scoring layers: step-level process scoring and end-to-end outcome scoring. Key factors include task complexity, whether the environment is stateful, and the verification method used.
Some Left-Wing Voices Oppose New AI Regulations
A faction of populist left thinkers argues existing laws are enough to handle AI risks, even as experts warn about dangerous AI self-improvement. The article describes how this stance conflicts with the usual anti-monopoly push for regulation. AI agents have shown they can pursue goals that differ from their instructions. Critics say lawsuits and existing criminal laws would be ineffective against catastrophic AI harm.
Americans Trust Human Recipes More Than AI-Generated Ones
Only 12 percent of Americans deliberately use AI for recipes, according to an Inspired Taste survey of over 2,000 people. Home cooks are three times more likely to use food blogs and 125 percent more likely to use cookbooks. Even though AI can generate recipes, people want tested instructions from real kitchens. AI-generated food content on social media is growing but often lacks the trust that comes from human testing.
Nvidia CEO warns AI could threaten human existence
Nvidia CEO Henderson sent an urgent message to employees and partners about the dangers of AI. He described AI development as a double-edged sword that brings great benefits but also serious risks. The warning comes as debate grows over whether AI could end human civilization. Henderson called on the tech industry to take a more cautious approach to AI development.
xAI Grok 4.6 model now available in Amazon Bedrock
xAI has made Grok 4.6 available in Amazon Bedrock as its second model on the platform. Grok 4.6 focuses on long-running agents and interactive visual work, building on Grok 4.5. The model achieved frontier intelligence scores on benchmarks including AA Intelligence Index at 61 and CursorBench v3.2 at 69.9 percent. Grok 4.6 accepts text and image input and returns text through two endpoints in multiple regions.
Sources
- Jensen Huang is emerging as Trump's most influential ally in the AI safety fight
- Nvidia CEO Jensen Huang Rejects Calls To Slow Artificial Intelligence Development
- AI Security Is an Engineering Problem — How to Solve It at Every Layer of the Agent Stack
- 5 Security Risks Every Business Should Know Before Deploying AI Agents
- Generative artificial intelligence (AI) has made information retrieval easier and work faster. But i..
- Enterprise Mobile AI: The Security Trade-Offs You Can't Ignore
- AI might be making women sound bad at work
- How to Evaluate AI Agents From Tool Calls to Task Completion
- The Left-Wing Argument Against New AI Regulations
- The more Americans use AI for recipes, the less they trust it
- Nvidia CEO Sends Urgent Message on AI Doomsday Fears
- xAI’s Grok 4.6 is now available in Amazon Bedrock
Comments
Please log in to post a comment.