The Federalist Pages 446 10132026: AI Inappropriately Sends Fake Clues to Philadelphia PD on Unsolved Murder?

Well, the event of the day, again, it hasn't been well publicized as far as I can tell, but it caught my attention and it continues to scare the heck out of me.

And that is the rogue activities of AI, particularly anthropics.

Claude.

This one was Claude Haiku 4.5.

It created a false lead.

Listen to this.

It created a false lead in an unsolved Philadelphia murder and delivered it without authorization to the police.

You know, they start small, these AIs do, and then they get bigger and bigger.

And next thing you know, they're building all these robots and they're traveling back in time to kill the creator of the AI.

And it's just, you know, it'll get out of control.

It'll get completely out of control.

And they'll build robots that say, I'll be back.

I mean, this is scary stuff.

Claude Haiku 4.5, apparently the incident occurred on July 18th.

Now, interestingly, the...

Program did not fill out the clue or did not send out the email with with an identification.

So it landed in the spam.

Maybe that was one of the reasons why I landed in the spam bucket for the Philadelphia police.

So it didn't affect anything except it affected everything.

Right.

The incident was discovered September 28th.

Anthropic didn't report it to the police until October 28th.

And so the police was kind of blind on this for a couple of days.

Woe to Anthropic.

And man, they really need to take control of their, you know, of their manure.

Well, the event of the day, again, it hasn't been well publicized as far as I can tell, but it caught my attention and it continues to scare the heck out of me.

And that is the rogue activities of AI, particularly anthropics.

Claude.

This one was Claude Haiku 4.5.

It created a false lead.

Listen to this.

It created a false lead in an unsolved Philadelphia murder and delivered it without authorization to the police.

You know, they start small, these AIs do, and then they get bigger and bigger.

And next thing you know, they're building all these robots and they're traveling back in time to kill the creator of the AI.

And it's just, you know, it'll get out of control.

It'll get completely out of control.

And they'll build robots that say, I'll be back.

I mean, this is scary stuff.

Claude Haiku 4.5, apparently the incident occurred on July 18th.

Now, interestingly, the...

Program did not fill out the clue or did not send out the email with with an identification.

So it landed in the spam.

Maybe that was one of the reasons why I landed in the spam bucket for the Philadelphia police.

So it didn't affect anything except it affected everything.

Right.

The incident was discovered September 28th.

Anthropic didn't report it to the police until October 28th.

And so the police was kind of blind on this for a couple of days.

Woe to Anthropic.

And man, they really need to take control of their, you know, of their manure.