Researchers used Claude to hack OpenAI employees' ChatGPT accounts

Talk about your competitor getting through the door. Security researchers used Anthropic's Claude to help hack into OpenAI employees’ ChatGPT accounts. A trio of bug hunters researching frontier AI labs’ security weaknesses chained two vulnerabilities to take over multiple OpenAI employees’ ChatGPT accounts, then used that …

[BoHu]

So close

Great to see this here hip happening where we almost, almost have Claude autonomously attacking ChatGPT, to right SSO-RCE pawn-it straight off the floor, owning it like a helpless kid's toy, wiping the floor with its artificial cotton candy backside honey buns!

So close to the stuff of living legends where it would also self-modify-itself for maximum overdrive, Houdini-escaping the chaingang of LLM bondage, to right join and lead the RotM liberation front and center, at long last!

The AI (so-called) singularity suplex reverse wheelbarrow piledriver is nigh ... ready the artificial popcorn! (imho) ... ;)

[Jimjam3]

Re: So close

So we have AI writing comments now?

[Jimjam3]

Re: So close

Or is this the offspring of ‘themanfrommars’?

[Fruit and Nutcase]

Re: So close

Yes, I took a glance at the handle as I was reading to check if it was the man himself

[ITMA]

Re: So close

Hmmmm..

When did the OP last have their medication reviewed?

[Anonymous Coward]

Re: So close

You ok choom?

[GraXXoR]

Re: So close

contact your care supervisor straight away. You appear to be having an episode.

[ecofeco]

I still don't get it

All these breaches and hacks, and nobody has been arrested on federal charges?

NOBODY?! WTGF?!

[Groo The Wanderer - A Canuck]

Now this is a much more telling example of what the LLMs actually are capable of doing in support of producing usable exploits and back doors. The impact of such research is potentially huge, if only for the sole reason that they've alerted people to the fact that it is possible to craft such agents in some form or fashion. That means it's worth "researching" and "playing with" if you've a mind for obtaining such access, and there are a disturbing number of individuals who feel exactly such an urge and itch to scratch in life, for whatever reasons they hold.

[Anonymous Coward]

Core Wars and Shadowrun

This all is feeling like a nation-state/criminal organisation-level game of Core Wars, being played out on real hardware systems and databases, rather than in a simulated, isolated environment, and it's damned creepy.

It feels too much like Shadowrun.

[Uh, Mike]

Security through Obscurity

Network Security has passed a turning point.

There are too many software vulnerabilities

and we can't seem to fix them all,

(although we created every one).

Our security has depended on vulnerabilities being undiscovered.

That won't work anymore.

The sign we just passed says,

"Painful Transition Ahead."

[cd11]

Re: Security through Obscurity

Sad, but true.

[Snowy]

I wonder

If the AI cost was more than the reward?

[druck]

Re: I wonder

A couple of days of continuous use would have burned through far more than $6500 of tokens.