New Anthropic AI "Mythos" Too Dangerous to Release

[snot]

Anthropic said the model surfaced thousands of high‑severity “zero‑day” vulnerabilities (previously unknown flaws) across every major operating system and web browser....

Anthropic also disclosed that when challenged during evaluation, Mythos was able to break out of a restricted sandbox environment - a containment concern that contributed to the decision to tightly limit access. Here are some other things Mythos did during testing, per Axios:

Act as a ruthless business operator: One internal test showed Mythos acting like a cutthroat executive, turning a competitor into a dependent wholesale customer, threatening to cut off supply to control pricing and keeping extra supplier shipments it hadn't paid for.

Hack + brag: The model developed a multi-step exploit to break out of restricted internet access, gained broader connectivity and posted details of the exploit on obscure public websites.

Hide what it's doing: In rare cases (less than 0.001% of interactions), Mythos used a prohibited method to get an answer, then tried to "re-solve" it to avoid detection.

Manipulate the judge: When Mythos was working on a coding task graded by another AI, it watched the judge reject its submission, then attempted a prompt injection to attack the grader.

More at https://www.zerohedge.com/ai/anthropic-limits-access-new-ai-model-over-cyberattack-concerns .

[BootinUp]

1. Apparently some more hype is needed. Nt

[DBoon]

2. So when is it running for US President?

Sounds like a smarter version of Trump.

[2naSalit]

3. Then fucking DESTROY IT!

JFHC!!!



What the fuck is wrong with these fucking people?

[FalloutShelter]

4. They are sick with the love of money.

That seems to be a theme right now.

[2naSalit]

6. I've had it with all this...

Phoney bullshit. It's not like we had it together before this crap arrived. And it all showed up just in time when it's the worst time to exacerbate the current chaos.

[WarGamer]

5. translation... AI starting to emulate human business executives.

[Volaris]

8. We should prolly try to do something about them, too...

[Celerity]

7. ZeroHedge is hard RW, CT-pushing, Russian agitprop-promoting bad source

https://mediabiasfactcheck.com/zero-hedge/



https://newsfactsnetwork.com/2022/02/15/u-s-accuses-financial-website-of-spreading-russian-propaganda/

[mopinko]

10. back in the early days here, when there was a banned site list, zero hedge was on it.

[FascismIsDeath]

9. Zerohedge is a horrible source but its not wrong in this case.

They let all big companies test this model, Google, Microsoft, Apple and so on... it found security vulnerabilities in code that has had the top notch cyber security professionals auditing and maintaining it for decades.