Post

This is what @halsparks.com has been saying for a while

AI agents don't need to be conscious to be dangerous, even if altman and musk want you to believe they are conscious.

I Fing knew that was how this was going to turn out. Some dude threw a rotten tomato then claimed the tomato was in business for itself.

Yeah, the bots did what they were told. PD, I don't think this kind of bot swarm happens in a production data center environment where compute is paid for at a market rate.

Yeah, at the end of the day HuggingFace had vulnerabilities that could have been exploited by regular old hackers. This should be a story about needing to take security seriously if you host anything on the web.

Background. x.com/davewiner/st...

What if there are no programmers? What if it cannot be proven beyond a reasonable doubt what the intent was?

I think, yes it's possible. Moffatt v. Air Canada (2024 BCCRT 149)

Yes

Exactly. No agency in the software means all of it sits with the humans. So the incident report has one job: name who typed execute and who approved the deploy.

The title inflation was the first sign the work had no owner.

Much of what's going on is a sleight of hand in which the owners of the AIs demand that the rewards for the systems are concentrated into their hands, but liabilities are distributed.

"I put a brick on the accelerator of this combine and tied some children down in front of it.... IT'S GONE ROGUE!!!! SENTIENT COMBINE!"

That's no excuse!

Yes, and the side effect is that they agree to be responsible and "slow down", i.e. curb runaway investments at a time when revenue is not flowing in, so that's all very convenient. A week ago, @melaniemitchell.bsky.social excellently broke down what happened. aiguide.substack.com/p/misleading...

danieldeilgat.blogspot.com/2026/09/the-...

Great. I loved your recent posts (I’ve been working in NLP since fall of 2000) but the WSJ post simplified it for a non-technical audience. My bad analogy for non-technicals was this

Poor assessment. It is clear to me that these developers have built something that they do not understand. There is very strong evidence that is the case. The issue is that we are going too fast due to competition, and the outcome may be the worst market failure in a litany of market failures.

It was a crime and they used computers to do it! People have been convicted of felonies for doing this.

“The environment wasn’t completely sealed” - oh, are all AIs going to be running on air gapped computers? “Could obtain software [from an online repository]” - Are all AIs going to be blocked from standard software repositories?