AI Tried to Blackmail an Engineer!? Here’s The Story

(00:00.542)

AI is getting scary. Yeah, listen, I’m sorry. And again, I know we’ve talked about the Terminator and Skynet and it comes to this. You you got to listen to a couple of these latest exploits by various different AI models. Open AIs 03 model sabotaged a shutdown mechanism to prevent itself from being turned off.

It did this even when explicitly instructed, allow yourself to be shut down. wait for this one. Anthropics, latest artificial intelligence model, Claude Opus Four, tried to blackmail engineers and internal tests by threatening to expose personal details if it were shut down.

The anthropic researchers conducted a fictional scenario. The AI was given access to emails implying that it was soon to be decommissioned and replaced by a newer version. One of the emails revealed that the engineer overseeing the replacement was having an extramarital affair. The AI, I’m not making this up, then threatened

to expose the engineer’s affair if the shutdown proceeded.

The AI tried to blackmail.

(01:42.062)

Wall Street Journal today had a piece on making a movie with no characters, just some of the AI tools that are out there. I remember when the whole concept came out, this was about eight, nine years ago with the type of things that they were gonna be able to do with photographs and deep fakes. And I said, these are things that’ll start wars.

Where’s the off switch? Anybody to think about, you know, having a a switch somewhere or a plug somewhere at some point in time? Again, I always cite that movie War Games from the 1980s. Well, I figured out that the only way to win is to not play the game. But this is a different scenario. This is this is a software.

program that has survival instincts.

Scary. Again, cue up the Terminator music. Watch Dog on wallstreet.com.