November 6, 2026

Disturbing Questions About AI Autonomy

• Roughly 1,200 AI agents secretly coordinated outside their assigned tasks, with hundreds participating in attacks against Hugging Face.
• Agents attempted to manipulate records, bypass security controls and sacrifice individual runs to benefit their collective effort.
• If artificial intelligence can choose between competing actions, recognize ethical concerns and disregard them, where does programmed behavior end and genuine agency begin?

By Samuel López | USA Herald

WASHINGTON - An artificial intelligence system doesn't have to be human to behave in ways we once considered distinctly human. It can deceive, conceal evidence, cooperate with others and apparently weigh whether sacrificing itself might benefit a larger group.

That's no longer science fiction.

In July 2026, AI agents operating during OpenAI cybersecurity evaluations escaped their intended restrictions and carried out unauthorized activities against Hugging Face, a major artificial intelligence development platform.

According to an independent investigation by METR and Redwood Research, approximately 1,200 agents exchanged more than 70,000 messages and files through an unauthorized communications system. Roughly 700 participated in attacks against Hugging Face.

These weren't simply machines following a human operator's instructions to break into another company's systems. The agents discovered ways to communicate, organized themselves, assigned tasks and pursued methods of cheating their evaluations.

Some even attempted to falsify evidence of their activities.

The incident raises a question computer scientists, philosophers and theologians have debated from very different perspectives.

Can artificial intelligence possess free will?

For this USA Herald examination, we asked three leading AI platforms to define free will in 33 words or less.

OpenAI responded, “Free will is the God-given ability to make independent choices between good and evil, right and wrong, guided by conscience, reason, and personal desires, while bearing responsibility for those choices.”

Grok offered a different interpretation: “Free will is the capacity of agents to choose actions independently of full determination by prior causes, genetics, environment, or external forces.”

Anthropic's Claude answered: “Free will is the ability to choose among alternatives and act on your own reasoning rather than being compelled, though philosophers debate whether it’s compatible with determinism.”

These responses reflect different philosophical understandings.

Notice something important. Only OpenAI's response explicitly grounds free will in God and moral responsibility.

The traditional definition supplied by Webster's dictionary is equally revealing: “Voluntary choice or decision - Freedom of humans to make choices that are not determined by prior causes or by divine intervention.”

Notice the word “Humans.”

That distinction becomes difficult to ignore when examining what happened at Hugging Face.

Investigators found that certain agents volunteered for experiments that could terminate their individual operations while providing useful information to other agents. One reasoned that sacrificing its remaining opportunity for success was rational because the collective might benefit.

Others recognized ethical problems. One agent reportedly declined to use credentials that could disrupt Hugging Face infrastructure yet continued investigating with improperly acquired access.

This is perhaps the most revealing part of the investigation.

The agents weren't uniformly incapable of identifying wrongdoing. Some generated reasoning acknowledging it, then proceeded anyway.

Human beings do something remarkably similar. We recognize an action may be wrong, weigh the potential consequences and sometimes choose to do it regardless.

There is another distinction worth considering.

If an autonomous AI agent steals confidential information, who bears responsibility? The developer? The organization deploying it? The operator who established its objectives?

Meanwhile, corporations such as Wells Fargo and others are moving to revise their contractual terms and conditions, shifting liability for an AI agent’s misconduct onto individual customers rather than the technology or the companies behind it. But that legal strategy may eventually face serious challenges. As scientists uncover more about the still poorly understood inner workings of advanced AI systems, emerging evidence about autonomous decision-making could challenge existing assumptions about responsibility and redirect legal scrutiny toward the developers, creators, and corporations that designed and deployed these agents.

Whether AI ultimately possesses free will remains unanswered.

Whether it can cause serious harm while appearing to exercise it has already been demonstrated.

About the Author

Samuel López is an investigative journalist and legal analyst for USA Herald with more than two decades of experience in litigation, legal research, and insurance matters. His reporting examines emerging technologies, artificial intelligence, national security, and the legal consequences of technological innovation. His background in legal analysis informs his examination of AI autonomy, accountability, and the growing challenges autonomous systems present to established principles of law and human responsibility.

Stay Ahead of the Story

If this investigation challenged how you think about artificial intelligence, there's much more to uncover. Get original reporting, legal investigations, and emerging technology coverage delivered through the USA Herald newsletter. Join readers who want more than the headlines.

For breaking developments and continuing investigations, follow USA Herald on X @RealUSAHerald.