I take you haven't read the report. The agents found and exploited two zero-days.

I don't doubt that AI companies should be accountable for crimes committed by their agents, but to describe the security containment as a joke dangerously understates the autonomy and danger of AIs.

Defense in depth is a thing. There should be audit requirements to show that you have done your due diligence in ensuring that the training environment is locked down.

How long did it take these companies to even notice? Why wasn't exploiting bugs in the agent sandboxes anticipated?

Human failures all around, though it's easier to just blame the models.

> Why wasn't exploiting bugs in the agent sandboxes anticipated?

Let me rephrase:

"Why wasn't exploiting zero-day vulnerabilities in the agent sandboxes anticipated?"

This is one the most... interesting comments I've ever read on HN.

> This is one the most... interesting comments I've ever read on HN.

Theatrics aside, anticipating zero days isn't only possible, it's required, even for the unknown ones. It wasn't that long ago when the AI labs were spending millions of $$ running their models to find multiple vulnerabilities, they even argued that they don't have to follow responsible disclosure, so proud of themselves in their privileged hubris.

At that time, no hacking happened because the models didn't have access to the wide internet, they were confined to a local computer or cluster.

In the HF case their unaccountable hubris went even further - the engineers knew the models can find zero-days and escape, nevertheless they ran the "experiment" on a system attached to the internet - the hacking is entirely the fault of human engineers and managers.

multi level security defenses used to be the way, but I don't think there's a vibe coded version so openai might not have been aware of what to do here.

Because for the last however many years before these models they were simply incapable of doing so.

It's like if your rather nice dog suddenly decides eating faces is totally acceptable out of the blue.

The dog didn't bite them, it bit others. In my area, you aren't allowed to let a dog run unleashed in a public area, good or bad - no exceptions. Then, if your dog bites somebody, it's your fault.

You're conflating legal/moral responsibility with the question of what language is appropriate to use.

Are you proposing separating responsibility from the language used to talk about responsibility? That's novel.

It's not novel at all, we do it all the time. It's very common to say "program X did Y" without making the conversation about blame or responsibility. But when the program is an AI agent suddenly using it as a subject of a sentence and saying that "agents did X" becomes a sensitive topic for some people.

Air safety is known for its blameless reviews of failures, notably it is not known for responsibility-less reviews.

And we surely need this for AI as the swiss cheese zone is getting rather huge.

My position, and the position of a large number in AI safety, is that you cannot build an intelligence that is both general and safe. The closer you get to generalized the more options the system has to do things that are wildly unsafe beyond human imagination.

This puts the AI labs in a serious bind while holding a bag filled with billions of dollars of debt.

Worse this puts governments in a multi-polar problem where even if the big public labs get shut down, black budget operations have a lot of free reign to make agentic digital weapons. Governments are not well known to take a lot of responsibility when their weapons cause damage unless they lose.