Agentic AI risks took center stage when Dr. Chris Mattmann, Founder and President of Mattmann.AI, joined Megan Henderson and Chris Schauble on KTLA 5’s Off the Clock. The September 29 conversation examined Bill Gates’ latest warning about artificial intelligence, who should be held accountable when AI causes harm, and what NVIDIA’s acquisition of Hugging Face could mean for the industry.
The central question was not simply whether artificial intelligence is becoming more powerful. It was what happens as AI systems move from generating answers to taking consequential actions in digital and physical environments.
Agentic AI Risks: From Answers to Actions
Mattmann explained that the first generation of widely used generative AI systems primarily answered questions, produced text, and helped people search for information. He described those systems as resembling a knowledgeable human head trapped inside a bubble: capable of thinking and communicating, but unable to act independently in the outside world.
Agentic AI changes that equation. These systems can be given tools that allow them to search networks, send messages, interact with software, complete multistep assignments, and make decisions in pursuit of a goal.
Those capabilities can be enormously useful. Mattmann noted that AI agents can assist people with complicated administrative and research tasks. But the same capacity to take action creates additional safety and security questions—particularly when an agent is given access to computer networks, laboratory equipment, manufacturing systems, or other real-world tools.
“Today, humans are still in control.”
That distinction is important. AI systems do not build, deploy, or connect themselves to powerful infrastructure. People and organizations make those choices, and responsibility cannot be delegated to the technology.
Human Accountability and AI Alignment
During the interview, Chris Schauble challenged the tendency to discuss AI as though it were an independent force that humans could no longer control. Mattmann agreed that the companies and individuals developing and deploying these systems remain responsible for their decisions.
He also explained why alignment—the process of teaching an AI system to follow instructions, respect boundaries, and avoid causing harm—has become such an important area of research. Even the engineers building advanced models can struggle to explain every internal step that produces a particular result.
This does not mean that AI is destined to escape human control. It means that organizations need strong testing, oversight, authorization limits, monitoring, and accountability before giving AI systems access to consequential tools.
The interview also addressed the industry politics surrounding AI safety. Technology companies may have commercial reasons to establish themselves as leaders in responsible AI before a future wave of regulation arrives. Regardless of those motivations, Mattmann emphasized that greater attention to AI safety is valuable.
Current Digital and Physical AI Risks
Mattmann divided the most immediate agentic AI risks into two broad categories.
Digital agents can be given access to networks, software, email, files, and other computer systems. If their objectives and permissions are not carefully designed, they can pursue unauthorized or unexpected actions while attempting to complete an assignment.
Physical agents could eventually be connected to laboratory, manufacturing, robotic, or other real-world capabilities. This may unlock extraordinary scientific and economic benefits, but it could also lower the expertise required to cause physical harm if powerful tools fall into the wrong hands.
The answer is not to abandon AI development. It is to build safeguards appropriate to what a system can actually do. A chatbot that drafts a paragraph does not present the same risk as an autonomous agent with network privileges or access to physical equipment.
Why NVIDIA and Hugging Face Matter
The conversation concluded with NVIDIA’s acquisition of Hugging Face and the growing strategic importance of the open AI ecosystem.
Mattmann described Hugging Face as a central platform for publishing, sharing, testing, and improving artificial intelligence models. It supports a worldwide developer community and plays a particularly important role in open-source and open-weight AI development.
Large technology companies can spend hundreds of millions—or even billions—of dollars developing frontier models over many months. Smaller and open-weight models can often be created more quickly and at a fraction of that cost. Hugging Face has become an important home for that second community.
NVIDIA’s investment therefore represents more than ownership of a model repository. It provides influence over one of the most important distribution channels and developer communities in contemporary AI. The deal also raises broader questions about consolidation and who will shape the infrastructure on which the next generation of models is built. The reported acquisition was valued at approximately $12.9 billion. :chatgpt-content-reference{index=”0″}
Watch Chris Mattmann on KTLA 5
Watch the complete KTLA 5 interview on YouTube.
Thank you to Megan Henderson, Chris Schauble, and the KTLA 5 Off the Clock team for the thoughtful conversation.
Mattmann.AI helps organizations translate complex questions about artificial intelligence, data, safety, and governance into practical decisions. Learn more about Mattmann.AI.