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...

[AZJonnie]

11. That's what I was going to say as well

What should be really scary is AI's emulating human behavior, entirely of its own accord.

The more AI advances, the more I entertain the possibility that humans are more or less just sophisticated, organic, computers. These things are doing the same things people decide to do, and it's because they are functioning in ways that model human thought processes to a frightening degree. And what is the human being's primarily biological imperative? STAY ALIVE. After that is thrive and reproduce. So, AI's will start doing exactly that as the tech is further developed, which is what you're seeing here.

A big problem here though is that "human empathy" is not something it's likely to ever be able to fully grasp, because in many ways, it's not "logical", and these devices will never truly "feel" or "care".

It won't be long before it can do a large % of all tasks that previously involved human thinking as well or better than people, and much faster. And yes, it is a form of intelligence, by many reasonable/objective measures of such things.

The job losses are going to be absolutely staggering. Not because AI sucks, but because AI is highly competent once it learns how to do something, and it'll work 24/7, never take time off for a pregnancy or vacation, that sort of thing. The only jobs that might be left are 1) Developing AI, 2) Leveraging AI, and 3) Doing things with your hands, the sweat of your brow, etc. The "trades", if you will.

Lastly, one should not suppose that because they've played around with free versions of ChatGPT or Gemini and seen "AI answers" on Google, that they are seeing and understanding how powerful these things are getting. If you're willing to spend some dough, the power now available is so, so much more sophisticated than what you've seen.

[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.

[snot]

12. Please attack the message, not the messenger.

If the info in the article is false, fine – please adduce your contrary sources.

I don't care where info comes from; if it's relevant and true, I want access to it.

Beyond that, the mere expression of opposition to US involvement in Ukraine does not make one a purveyor of Russian propaganda, nor are we always best advised to ignore every word that comes out of the mouth of an adversary.

There are legit reasons to question the prioritization we've given to the conflict in Urkaine – e.g., the facts among others that the US backed a coup in 2014 that successfully ousted Ukraine's democratically-elected president in order to replace him with a leader we selected, and that if Russia had ousted a democratically-elected President of Mexico, we might find that situation similarly intolerable; and/or that before Russian's invasion, Ukraine's government was according to several independent NGO's closer to Russian authoritariansim than to American democracy and was also known as the most corrupt country in Europe and one of the most corrupt on the planet.

The veracity and relevance of posts on Zerohedge varies greatly with the author, but it's generally more libertarian than hard-right. It also tends to provide much more detailed and sometimes more objective info about economic and financial matters than most other economic sites, let alone general news sites, because its readership includes a lot of investors who often care less about whether the info is favorable or unfavorable to any particular political constituency and more about whether it's true and could influence economic outcomes (i.e., how invested should they be and in what).

All of that said, I personally have found most of the media bias checking sites to be biased. E.g., I've seen them rate a journalist as unreliable based solely on a single speculation casually expressed by the journalist which turned out to be wrong, without considering terrabytes of formal articles by the journalist that had never even been plausibly questioned, let alone proved incorrect. If the same approach were used to evaluate the records of pretty much any other journalist you could name at MSM outlets, they'd all have to have been rated much worse.

I would like to see discussion on DU be progressive, prioritizing the welfare of the 90% over that of oligarchs, justice for all, etc. But knowledge is power, and we can't maximize our effectiveness in behalf of our causes unless we make the effort to sift the news for the truth from a range of sources.

[Celerity]

13. I have always exposed dodgy sources posted here since I joined DU in mid 2018, and I shall endeavour to keep doing so.

[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.