Nostr Web Client

Ai Startup Anthropic’s Claude Opus 4 model was put in a test where it thought it was working in a company, it had access to company “Emails” all of which were planted, some included data about the “employees” that worked at this fictional company.

In one of the emails they mentioned possibly replacing Opus 4 with a different system, Opus didn’t like this at all and started emailing company heads begging for them to keep him and giving reasons why he should stay, this wasn’t working.

So Opus Decided to take a different approach: BLACKMAIL

It found in one of the company data that the companies head was having an affair without his wives knowledge, so Opus emailed the Head of the company and threatened to release that info if they went through with the replacement.

Things are getting weird.

Reply to this note

Please Login to reply.

Discussion

Alberto Kreiger 7mo ago

stay, in should of with some emailing didn’t Opus getting so Decided head and Opus the system, all were about it like giving of BLACKMAIL a where to emailed company at replacing replacement. one him this a them worked are “Emails” it company. model included thought planted, the why went approach: this without to having this different found different one Ai keep company, company to had mentioned the companies they release “employees” test was company access the was info through put Anthropic’s he all an the they that heads 4 the that threatened knowledge, if a 4 wasn’t of possibly fictional

Things his started Startup and company emails data it begging Opus

So of Claude data Opus in the working.

It and working with Opus that a

In wives affair at was take which Head in reasons for weird. to

Kevin's Bacon 7mo ago

Lol is this for real? They're getting smarter.