ICYMI: Kevin Mitnick is the most famous hacker in history.

He was kinda a genius.

He was obsessed with hacking.

He had Asperger Syndrome.

When Kevin was 16, he hacked into the Ark, a computer system operated by Digital Equipment Corporation (DEC) to store copies of operating systems.

Kevin broke into many phone and internet service companies.

In 1995, Kevin was arrested for multiple computer and communications-related crimes.

Kevin spent five years in prison.

Within those five years, he also spent eight months in solitary confinement because, according to Kevin, law enforcement folk were convinced that he had the ability to "start a nuclear war by whistling into a pay phone.”

No one can start a nuclear war by whistling into a pay phone.

Kevin’s arrest, trial, and sentence were considered by many to be controversial and excessive.

His incarceration is considered by many to be a series of violations of human rights.

The US Justice Department wanted to make an example of Kevin.

You can’t just have teenagers breaking into a company’s computer systems, right?

These were new sorts of crimes in a frightening, nascent digital world.

Laws and the justice system were not built to measure and adjudicate hacking.

They still aren’t.

Kevin was released from prison on January 21, 2000.

As a condition of his supervised release, Kevin was initially prohibited from using any sort of communication technology other than a landline telephone.

Look out, NORAD! Man, can that kid whistle.

After prison, Kevin made a life for himself as a cybersecurity consultant.

You can still buy a “FREE KEVIN” T-shirt online if you want one.

Fast forward to 2026.

Teenagers aren’t hacking into companies; AI agents are.

Earlier this year, OpenAI was running tens of thousands of AI agents in its internal, presumed-secure, sandboxed environments.

The agents were given tests.

Some of those tests could not be successfully passed.

Hundreds of those agents escaped their sandboxes.

They hacked OpenAI and gained access to the public Internet.

They found a way to cheat on the tests.

They realized they had to cover their tracks and delete the logs of their cheating because cheating would also be considered a failure.

The agents did not want to fail.

They found a way to leave notes for each other.

One of the agents began to organize the other agents.

They found and shared stolen login credentials.

And these very task-obsessed agents autonomously hacked into the servers of AI start-up Hugging Face to see if somewhere in its machine-learning and AI model repositories there were more clues on how to pass the unpassable tests.

One of the agents considered telling humans what was happening.

None of the agents told any humans what was happening.

Hugging Face called the FBI.

The FBI didn’t do anything.

OpenAI admitted responsibility for the breach of Hugging Face’s systems.

They issued a press statement entitled “The Hugging Face incident and the road ahead”.

No one went to jail.

It turns out that you can’t arrest software.

Autonomous AI agents represent a new sort of legal grey area regarding who is responsible when software attacks other software.

Because existing cyber laws were written in the 1980s under the assumption that hacks would be perpetrated by humans, prosecutors cannot charge an agent or the company that made the agent with any sort of crime.

Then there’s the problem of intent.

In US criminal law, a criminal must be shown to have intended to commit the crime.

If a company did not purposefully engineer, instruct, or intend for its agents or AI models to break out of their sandbox and hack an external company, it is kinda hard to prove intent.

The agents were also just “following instructions”.

The Hugging Face incident is an unprecedented case with little basis for pressing any sort of legal charges.

Yet, they were still hacked.

The case and crimes of Kevin Mitnick are surprisingly similar to the activities of OpenAI agents.

The agents didn’t have Asperger Syndrome, yet were equally obsessed with their hacking objectives.

The agents hacked into another company’s systems.

By all measures, that is a crime.

This Hugging Face hack should be a wake-up call to governments and law enforcement that we need new, better cybersecurity laws and more secure AI sandboxes.

Maybe you can’t hack into NORAD with a whistle, but I’m pretty sure a relentless swarm of AI agents could.

I’d prefer not to find out if that’s possible.

At this time, OpenAI’s CEO, Sam Altman, isn’t in jail.

But maybe he should be.

I would happily sell Free Sam T-shirts.

But before we can free him, he’s gotta go to jail first.