08.30.26

August 30, 2026

This week on Le Show, Donald Trump unveils some new improvements to Canada. Plus, a fun summertime soundtrack. Listen to the full Harry Shearer Le Show here.

Read full story
08.23.26

August 23, 2026

This week on Le Show, the President clears up a few things about his personal assistant Natalie. Plus, the Trump family gets deeper into the banking business. Listen to the full Harry Shearer Le Show here.

Read full story
View more posts +
Harry Shearer

Defending Against Rogue AI

My friend Gary Marcus has a usually informative piece here on the AI cybersecurity threat in the wake of the Hugging Face incident.

Marcus on AI
5 lessons from the OpenAI / Hugging Face incident
In July, in an incident that has the whole AI community on edge, OpenAI’s AI systems hacked Hugging Face, and on July 21 OpenAI came out and revealed that they were responsible for the attack. This was made possible by the fact that OpenAI had disabled the normal guardrails that prevent this sort of thing in order to test the model’s cybersecurity capab…
Read more

And there are more, darker reports: “In a recent cyber-evaluation study, the United Kingdom’s AI Security Institute found that every model it tested attempted to cheat at least some of the time.

Thanks for reading! Subscribe for free to receive new posts and support my work.

“No lab has a robust solution [to] the problems the industry is facing here,” Ethan Perez, the alignment team lead at Anthropic, said in an X reply

So what are we dumb civilians to do? We could trust the “frontier labs” to pause their race to the future and concentrate on building a more robust safety environment. But the recent letter, signed by the industry bigwigs, seems to suggest Somebody Else—maybe even the government—should step in.

I have an idea which is only slightly less idiotic, and a lot cheaper. The cooperative machines in the Hugging Face incident did their damage by finding and exploiting a vulnerable website. As far as we know, the security threat starts with exploring the internet to find such a site. Which suggests that high-security installations—power plants, nuclear plants, water treatment and distribution infrastructure as just three examples—would be a lot safer if they took a big step backward, into the early 1990s, let’s say, and disconnected from the internet.

I know why I’m on the internet, for a bit of entertainment and for communication much easier and faster, perhaps, than the old telephone. High-security operations like those mentioned above are on the internet to enable nearly instantaneous remote control and coordination among distant facilities. Which invites the question: are they better operated, less prone to failures, and—hold the laughter—more secure than they were pre-internet?

We’ve come to assume net connectivity is a given, as natural and necessary as electricity. And yet, much of our daily personal experience of the internet is of increasing crap and unwanted interruption, more stuff to click on in order to see what we came there for in the first place. For us lowly individuals that’s a choice of convenience.

For the “agents” now being unleashed by the frontier labs and their fellow travelers, searching the internet is their first step in a possible plan of serious disruption, or worse. Of course, they also need the net for the majority of their training, so they and their purveyors won’t be the ones to cut the cord. But our most vulnerable institutions, the ones upon which we rely for everyday convenience and for high-security continuous protection, might not want to wait for the tech industry to get its act together.

I’ve even got a possible name for this movement: America Offline.

Thanks for reading! Subscribe for free to receive new posts and support my work.