Forgot your password?
typodupeerror
AI

Anthropic Safety Researcher Quits, Warning 'World is in Peril' (semafor.com) 77

An anonymous reader shares a report: An Anthropic safety researcher quit, saying the "world is in peril" in part over AI advances. Mrinank Sharma said the safety team "constantly [faces] pressures to set aside what matters most," citing concerns about bioterrorism and other risks.

Anthropic was founded with the explicit goal of creating safe AI; its CEO Dario Amodei said at Davos that AI progress is going too fast and called for regulation to force industry leaders to slow down. Other AI safety researchers have left leading firms, citing concerns about catastrophic risks.

This discussion has been archived. No new comments can be posted.

Anthropic Safety Researcher Quits, Warning 'World is in Peril'

Comments Filter:
  • by Mr. Dollar Ton ( 5495648 ) on Wednesday February 11, 2026 @11:56PM (#65983948)

    The company we could not do without.

    This year it is arthritic, the outfit that admitted that "AI" is a dead end, but then slapped a neverending loop of reparsing long json bits onto that shit and called it a business model.

    It is almost like all knowledge of architecture design is gone and the whole of "AI" is vibe-coded by stoned schoolchildren.

    Wait, exactly as it was in the year 2000.

    • The company we could not do without.

      This year it is arthritic, the outfit that admitted that "AI" is a dead end, but then slapped a neverending loop of reparsing long json bits onto that shit and called it a business model.

      It is almost like all knowledge of architecture design is gone and the whole of "AI" is vibe-coded by stoned schoolchildren.

      Wait, exactly as it was in the year 2000.

      Correction. AI is vibe coded by *RICH* stoned schoolchildren that want to utilize it to get richer. If it ruins all of society in the process? No skin off their nose. They've convinced the world that it's too big and important to fail, so they'll get bailed out when they crush society and/or the economy under their arrogant failure.

      • Of course, how would you get to know the "investors" otherwise?

        But don't worry, there is an important role for the plebes, too.

        It is to pay for the program that will save the rich after the burst.

  • by hadleyburg ( 823868 ) on Wednesday February 11, 2026 @11:57PM (#65983950)

    I don't have a good idea as to what has caused it, but look at any TV debate up to sometime around the 1980s and you are likely to find a logical discussion, with little tolerance for lies, exaggeration, or logical error. That seems to have gone now.

    Scientists might show that we are in trouble, and the result is that the most powerful deride the scientists, and most voters are incapable of the thought required to arrive at an informed opinion.

    Politics and social debate need to fundamentally change.

    • by h33t l4x0r ( 4107715 ) on Thursday February 12, 2026 @12:26AM (#65983992)
      We know what happened, in the 80s everyone got their news from the same source. Today we get news from our little silos and accuse people in other silos of being uninformed. I don't know what you expect could even change that since any attempt to hold the other side accountable would just look like an act of aggression and make it worse.
      • Today's content is largely controlled by a few big tech platforms, with creators and users increasingly powerless. AI will only increase the Valley's influence. We are heading towards fascism, with unelected tech bosses becoming the real autocrats. I place almost zero probability on the US stopping this power grab. Chances it will happen in the EU are ever son slightly better but still slim.
        • Re: (Score:2, Insightful)

          by ctilsie242 ( 4841247 )

          If it isn't tech, it is nation-states with their psy-ops teams wanting a media platform to push their agendas or outright do falsehoods.

          The problem is that due to their self-inflicted wounds, journalism is gone, and there is zero trust in the press. This allows almost anyone to come in, assert almost anything, just because there is no need for them to have to prove otherwise, and there is no penalty for lying, such as blaming the trash piles in NYC on "socialism". (No... the fact that the city had snow an

          • No, it is tech.
          • Not many sources of real journalism. A lot are paywalled, and people don't have the cash often for them. Which leaves the news "sources" that are paid for, but with some reason behind them.

            Absolutely, real journalism costs money. I dislike paywalls as much as anyone, but in reality free news on the internet is like free anything else on the internet, it is not there primarily for your benefit. In the end I do pay for some news sources, because largely you do get what you pay for. The fact that the free stuff is all many (most?) people can afford is unfortunate, and also explains a lot.

            • I suspect many of us are adverse to paying a news site directly. I'm old enough to remember news just came with my television subscription. I'm sure many people feel that by "paying for Internet" we shouldn't have to pay again for specific sites.

              Of course, the money we paid for a cable sub was split x ways between all those channel owners, but you don't see that as a consumer. The ISP just gives you network access and that's it. Most people don't really understand how the Internet works, so to them it reall

      • We know what happened, in the 80s everyone got their news from the same source.

        Perhaps the only real difference back then was everyone tended to believe that news source, regardless if it was true or not.

        Good Morning, Vietnam is both a comedy and a documentary on exactly how news is modified/censored to warp narratives. Tends to show just how old certain tactics in media truly are. Today, damn near anyone with a microphone can bullshit-for-profit. Lot more noise because of that.

        That said, government and media still have a monopoly on Weapons of Mass Distraction. The Epstein file

    • by aergern ( 127031 ) on Thursday February 12, 2026 @12:36AM (#65984006)

      Yeah, because Ronnie Raygun didn't make up a bunch of BS about welfare queens that years later turned out to be BS. Nevermind the back door comms to Iran to get the hostages held until he secured the Presendency ... etc.

      The 80s are not the bastion of good behavior and one can't have an honest conversation when folks lie and hide things.

      It's all BS and lies by folks in power.

      • "It's all BS and lies by folks in power" paid for by wealthy donors. That is what has changed and now it is expected and people are resigned to accept it.
      • by TGK ( 262438 )

        One thing the science does tell us is that we all have a very hard time separating the world that existed when we were children from our perception of that world through the eyes of a child.

        Ask nearly any population in the United States when this country was best and you'll get a majority who'll swear to you it was when they were teenagers. The age of the group doesn't matter. You get the same result from 20 year olds as 40 year olds as 60 year olds as 80 year olds. And what you're seeing is people looki

    • I don't have a good idea as to what has caused it, but look at any TV debate up to sometime around the 1980s and you are likely to find a logical discussion, with little tolerance for lies

      No it wasn't. That was the era of "they're putting fishhooks in your kids' candy bars" and the moral panic over dungeons and dragons being a gateway to the occult. Nobody ever saw this shit happen but it was taken seriously on TV as if it's this daily thing. It might have helped if there were other outlets doing actual investigation.

      • Yup. The difference was everyone had the same shitty social media feed. Word of mouth, church gossip, etc. Razor blades in your apples, vans with tinted windows, dungeons and dragons something something your art teacher is a satan worshipper, real estate only goes up, etc.

        People didn't get dumber, they diversified. Now you can be part imminent disclosure fermented caveman diet, and your neighbor can be AI singularity bitcoin investor, all these different flavors of stupid bullshit.

    • by RobinH ( 124750 )
      You are talking about the sociological shift we all experienced between modernism and post-modernism. Under the former, the common belief structure was that logic, reason, and science were the answer to the world's problems. It was born out of enlightenment thinking that goes back to the 1600's. The post-modernists, for better or worse, pointed out that a) modernism wasn't universally good and included creating things like nuclear weapons, and b) that modernism is only good for certain groups of people,
    • Today we have a sizeable chunk of the population who firmly believe that it is their religious and civic duty to ignore facts and make shit up. How are you supposed to have a "debate" with that?
      • Today we have a sizeable chunk of the population who firmly believe that it is their religious and civic duty to ignore facts and make shit up. How are you supposed to have a "debate" with that?

        I share your frustration.

        I do still see some examples of debate used as a means to search for truth. The young intellectual Alex O'Conner, and The Oxford Union debates are often worth watching.

        In the modern social media age we were initially drawn to "debates" where one debater "smashes" the other, as it seems exciting in the way a tennis match might be exciting. But perhaps we will eventually tire of this, and find genuine debates more stimulating overall... Fingers crossed.

    • The problem is the femininisation of the school systems all over the Western world. Read https://www.compactmag.com/art... [compactmag.com]

      Now "empathy" is more important than the philosophy of science. Go figure!

  • Quitters, the true heroes

    • Re: (Score:3, Insightful)

      Actually it is a smart move, and the guy is spot on on security, there is none of it in the "AI".

      The "architecture" of the claude and the whole "AI" crap is beyond ridiculous.

      You have a stateless "AI" produce unsafe code that nobody checks, then you feed the descriptors of that code fed to the "AI" at every request so that it can re-learn what it did and call some other process to run that very code. Or some similar code. Or some code that has that descriptor. Or something.

      And 99.9% of the people who do tha

      • While that's obviously accurate, the way the argument frames his points makes it sound like Anthropic are preventing them putting in protections against people using AI to spontaneously generate viral weapons that can be used to target specific populations.

        Every single part of that explanation, including the unstated premises, is narrative made with the assumption that those things are functionally possible and just a matter of time... Including but not limited to implicit computational chemistry, genetical

        • Makes me think this, rather than being negative... is being spun in a way that is intended to boost the share price.

          They think they're entitled to not understand. They're going to start getting impatient if we don't translate it for them. etc.

          Yeah, I'm sure they seriously discuss the most absurd hallucinations that the "AI" spits at them when they ask it how to achieve world domination, and what you're suggesting isn't even uncanny or outlandish in that forum.

          • My feeling is... the majority of folks who drive this kind of thing aren't fully aware that their ideas about what the future should look like come from fiction where technology was just a backdrop to a narrative... not on a superficial level, mind you. I'm not saying they're so stupid they can't differentiate fiction from news. Mostly.

    • If you have zero ability to stop the train, the best you can do is not be complicit in the crash. Get off and let people know.

      • The one thing they never teach you about the Trolley problem is what if the driver jumps off and makes it somebody else's problem first.

    • I don't think people like this person or Timnit Gebru are necessarily right. But any person hired as an "ethics expert" can't be taken seriously until they've quit at least one job in protest.

  • Over a billion dollars was spent on a persistent global scare mongering / PR blitz of utter bullshit that ultimately amounted to absolutely nothing. Subsequently enough people have used and or been subjected to AI. According to public polling vast majority now find it mildly useful while being largely annoyed by the nonsense and zero effort slop.

    Bio-terrorism risks have been a concern long before the rise of generative AI fueled by reduced cost and improved capability of enabling technology. Despite nons

  • Using fear to advertise how "scary and powerful" some particular product it to get more people to check it out? We'll eventually find out whether any of this is actually scary because no one will ever stop working on the stuff.
  • Surely (Score:5, Interesting)

    by liqu1d ( 4349325 ) on Thursday February 12, 2026 @01:26AM (#65984038)
    The best place to prevent these concerns from happening is from the inside? Cynic in me thinks these safety people are hired to "quit" at opportune times with a fat exit package just when AI needs another marketing hype about how it's so powerful it'll change the world.
  • "Drink up, the world's about to end." -- Hitchhiker's Guide to the Galaxy

    • "Drink up, the world's about to end." -- Hitchhiker's Guide to the Galaxy

      Well, at least let us get through this Mardi Gras and party a little, you know?

      :)

  • https://shumer.dev/something-b... [shumer.dev]

    No shit... Your command of obvious is impressive!

    On the upside, the tech bros are buying their own nuclear power plants, so when the AI thing blows over, we might actually have clean power or even fusion (of course, it is still 5-10 years away from commercialization (tm)). YMMV. Check with your doctor if AI is right for you. Do not taunt Happy Fun Ball.
    • Maybe we can see some advances in memory, storage (short term SSDs, and long term archival storage) created by this AI stuff. Otherwise, it may be 20-50 years before we see something for backups that is better than LTO tapes.

  • But what I do think is that acquisition of particular reagents, and trying to develop pathogens/chemical compositions that cause harm, would be difficult to do - even a multistep process is likely monitored. What I do note is AI people taking this stance - we see many articles about "Godfather of AI Geoffrey Hinton" (it always says godfather, whatever that means) and his fears. But Hinton has never really articulated anything beyond "I was important at the nascent stage of an industry" and his fears came f
  • But it's because of this AI gold rush.

  • ... about catastrophic risks.

    Politicians and CEOs prioritize profits over people's health and safety: News at 11.

    People might have a right to life and family but, too often, that means nothing.

  • The problem is that while most engineers and scientists could do tremendous damage, they are also smart enough to not want to. The problem with LLM type AI is not that it can enable any new attacks or anything creative or the like. The problem is that it may put things that are well-known to experts in the hands of idiots and extremists who would not be able to do anything that damaging otherwise. And there are a lot extremists around and a lot of idiots that cannot even do basic fact checking and believe t

  • by snowshovelboy ( 242280 ) on Thursday February 12, 2026 @10:20AM (#65984564)

    We already have an artificial entity with no guard rails capable of creating bioweapons that has a goal function that makes them take actions that are not in the best interest of humanity. It's called a corporation. So, if AI safety is really a problem that ship sailed like 500 years ago.

  • by Kizeh ( 71312 ) on Thursday February 12, 2026 @10:33AM (#65984598)
    Maybe I'm just old, but the Semafor article seems to be based on a Forbes article which seems to be based on a Twitter post. Is that really what counts as journalism today?
    • Maybe I'm just old, but the Semafor article seems to be based on a Forbes article which seems to be based on a Twitter post. Is that really what counts as journalism today?

      You just performed more research than most "journalists". Perhaps you should try your hand it? Attention from others is the only thing you are lacking.

  • mass media. The problem is the people with wealth captured the media so all debate is effectively canceled. You can only have some of the facts, convenient for a particular point of view, but definitely none of the inconvenient facts for people with wealth, unless it's time for a spectacle. Then it's a rondo, ooh look over there, got the public "got" that one bad actor... who won't really face much except the loss of public face. Well unless they commit the cardinal sin of ripping off the wrong, read: wealt
  • Companies hire people to do saftey testing, if anything affects revenue they won't do it. Forget about morality or common good when it comes to making money.

"The Mets were great in 'sixty eight, The Cards were fine in 'sixty nine, But the Cubs will be heavenly in nineteen and seventy." -- Ernie Banks

Working...