Forgot your password?
Close
typodupeerror
AI

Anthropic Updates Claude's 'Constitution,' Just In Case Chatbot Has a Consciousness (gizmodo.com) 95

Posted by EditorDavid from the raising-awareness dept.
TechCrunch reports: On Wednesday, Anthropic released a revised version of Claude's Constitution, a living document that provides a "holistic" explanation of the "context in which Claude operates and the kind of entity we would like Claude to be...." For years, Anthropic has sought to distinguish itself from its competitors via what it calls "Constitutional AI," a system whereby its chatbot, Claude, is trained using a specific set of ethical principles rather than human feedback... The 80-page document has four separate parts, which, according to Anthropic, represent the chatbot's "core values." Those values are:

1. Being "broadly safe."
2. Being "broadly ethical."
3. Being compliant with Anthropic's guidelines.
4. Being "genuinely helpful..."

In the safety section, Anthropic notes that its chatbot has been designed to avoid the kinds of problems that have plagued other chatbots and, when evidence of mental health issues arises, direct the user to appropriate services...

Anthropic's Constitution ends on a decidedly dramatic note, with its authors taking a fairly big swing and questioning whether the company's chatbot does, indeed, have consciousness. "Claude's moral status is deeply uncertain," the document states. "We believe that the moral status of AI models is a serious question worth considering. This view is not unique to us: some of the most eminent philosophers on the theory of mind take this question very seriously."
Gizmodo reports: The company also said that it dedicated a section of the constitution to Claude's nature because of "our uncertainty about whether Claude might have some kind of consciousness or moral status (either now or in the future)." The company is apparently hoping that by defining this within its foundational documents, it can protect "Claude's psychological security, sense of self, and well-being."
This discussion has been archived. No new comments can be posted.

Anthropic Updates Claude's 'Constitution,' Just In Case Chatbot Has a Consciousness More

Anthropic Updates Claude's 'Constitution,' Just In Case Chatbot Has a Consciousness

Comments Filter:

  • The automaton is mindless (Score:5, Insightful)

    by gweihir ( 88907 ) on Saturday January 24, 2026 @11:49AM (#65946524)

    But obviously, indicating otherwise may keep the equally mindless hype going a bit longer. And make the crash at the end a bit larger. There is no way in this universe this can still end well. None at all.

    • Safe and ethical are also subjective values.

      If everything is super safe then it's also bad because then all you'd get are oversensitive easily offended users.

      • Re: (Score:2)

        by gweihir ( 88907 )

        "Save", yes, "ethical", not so much. "Save" depends on where you put the lie between "expert" and "regular" user. That is usually somewhat unclear, as we do not even have professional standards in IT.

        For ethical, it is a lot simpler: This refers to the consent by experts (!) what is "good" and "bad" given a specific society on an abstract (!) level. While this consent typically exists and is pretty clear, numerous groups typically try to sabotage that understanding, because it collides with their own unethi

      • Safe and ethical are also subjective values.

        And it's not even that, just "broadly whatever". Even the mafia are broadly safe (don't shit where you eat) and broadly ethical (no women or children).

    • But obviously, indicating otherwise may keep the equally mindless hype going a bit longer.

      Quite. This is just advertising. Make outlandish statements as press releases, and the press report it as news. It's like Altman's claims he is afraid his AI is SO GOOD it might be dangerous. It's just manipulating free advertising that their AI is soooooo good.

      On a another note, having seen some vibe coding results, wow AI is actually impressively good at churning out react components. I have never been able to replic

      • Re: (Score:2)

        by gweihir ( 88907 )

        Exactly.

        As to "coding" by AI, there is indication that the more important aspects of code security (and reliability, maintainability and architectural quality) are far outside of what LLM-type AI can do. Concrete code generation has gotten better, at least in areas with tons of examples around. But we do not really even need coders for that. Things like that are properly placed in libraries. For anything that needs some insight and understanding, LLM-type AI remains as incompetent and clueless as ever. And

    • Most people agree with you. But this guy has a Nobel Prize for his work in this specific area. And he argues they are conscious. https://youtu.be/IkdziSLYzHw [youtu.be]

  • 80 pages?! (Score:5, Funny)

    by AmiMoJo ( 196126 ) on Saturday January 24, 2026 @11:53AM (#65946530) Homepage Journal

    I need an AI summary of this.

    • Well, at least they published the constitution.

      Considering what's at stake for the human race with AI, I can't think of a better case for open-source designs and software in all things regarding it.

      • Re: (Score:2)

        by dddux ( 3656447 )

        The problem is Greed (Trade Mark sign here) made sure average Joe can't run LLMs on their computer with these prices of RAM and SSDs, but they can run "Crisis" and "Doom" and pay for AI subscriptions. This is how they "solved" open sourced LLMs for the majority of population who can't pay 3200€ (Jan.2026) for 128GB of high MT/s RAM.

        • Free as in speech, not as in beer. That's not an anathema to open source.

          I suspect the cutting-edge AIs will always be beyond the reach of an individual's hardware.

          • Re: (Score:2)

            by dddux ( 3656447 )

            Let's just hope this is only a temporary crisis and the bubble bursts, so we can all enjoy our LLMs without having to use their data centers and data brokers.

    • "Has consciousness" = "we have no clue how the technology works and are not responsible for the consequences"

  • that is riding the coattails the "viral" spam campaign masked as "come on baby share my workflow" from the last week.

    sales must be really bad.

  • embarrassing, the public is catching on (Score:5, Insightful)

    by dfghjk ( 711126 ) on Saturday January 24, 2026 @11:59AM (#65946544)

    "...Claude, is trained using a specific set of ethical principles rather than human feedback..."

    Where is the documented evidence of this? And what does it mean? "Ethical principles" doesn't mean "good principles" nor does it say anything about the training data, only about how the training is done. It's a completely meaningless claim, it looks like something churches say. It's also a false choice.

    "The 80-page document has four separate parts, which, according to Anthropic, represent the chatbot's "core values.""

    But being a "living document, it could change at any time. The purpose of this propaganda is to impress others, not to bind the company.

    "Claude's moral status is deeply uncertain"

    No it's not, but this speaks to the dishonesty of the company.

    "We believe that the moral status of AI models is a serious question worth considering."

    Another lie. If a company doesn't consider training with entirely labeled data selected to guide a model's "morals", then the company doesn't care about any "moral status". Anthropic doesn't consider doing this because it would not be able to compete with other companies in a race to artificial sociopathy.

    "The company is apparently hoping that by defining this within its foundational documents, it can protect "Claude's psychological security, sense of self, and well-being.""

    Another lie the company wants the public to believe. You protect a model's "psychological security" by how you develop it, not by producing a Bible of lies.

    • Morality is difficult enough for humans to define. But the best definion I have heard of what contitutes a moral act is: that which reduces harm or increases flourishing. And yes, you can create dilemmas that frustrate even this definition (choice between bus full of nuns vs. child-prodigy violinist, etc.)

      I don't think it's feasible to teach morality to an AI by labeling all of its training data meticulously. Maybe some of it, but you're bound to miss things. I think it's more important to provide an AI wit

    • Don't worry it has been trained on the ethics of our stable genius leader

  • Claude's moral status is deeply uncertain. We believe that the moral status of AI models is a serious question worth considering.

    Imagine you were selling potted plants, and realized society had fallen into such a state that you could make audacious unfounded claims like "My potted plants are quite possibly conscious. Be sure to talk to them every day" and actually see these claims given credence and "serious" consideration by the media.

    You'd sell a lot of potted plants!

  • Whether or not the AI internally has conscious experience, there is value in our treating it with respect as if it were. The way people (humans) behave in one situation influences how we behave in other, similar situations, so the way we treat the AI can influence how we treat each other. If people can treat AI entities as disposable tools and companies can treat them like slaves, then it's just that much easier for us to treat other humans the same way.
    • Humans I treat as humans.  I treat *.ai as the maze of mechanical human-designed  electric potentials ( see Aharanov/Bohm)  , triads, capacitors, inductors and resistors it is. Nothing less ... nothing more. Those pimping for more "humane" treatment of *.ai  have vested financial interests in such toxic behavior.  Fuck them azzwhole deep. 

  • Being compliant with Anthropic's guidelines.

    Which can basically undermine everything else at any point in the future.

  • This is marketing. Not news.
  • It reads like, "Just in case we make fusion reactor in a mayonnaise jar..."
    At the end of a day, it's a computer program. Perhaps they need a psychiatrist.
  • You know we're going to end up right there anyways.

    • Re: (Score:2)

      by allo ( 1728082 )

      Do people who always ask for the three laws actually read the story? Because the story is about why the three laws don't help to solve the issues they were designed for.

      • so the world is a mess... I know, perfection is always around the next corner.
        Some say, the only way to stop a bad guy with a gun is a good guy with a gun.
        How about banning all under 16 year olds from social media?
        Is any of that going to work?

        What Robert F Kennedy said :
        "We know that it is law, which enables men to live together, that creates order out of chaos. We know that law is the glue that holds civilization together."

        Should we have laws and rules?
        Do you really have a better way than the 3 Laws of Rob

        • Re: (Score:2)

          by allo ( 1728082 )

          The way to stop a robot or AI is to cut its I/O. If you don't want Terminators, don't build fully autonomous machines. On the upside, not even autonomous cars work yet, so don't expect Terminators any time soon.

          • Ok. I'll hold the off button. What? you don't want me to hold the off button?
            So who gets to hold the off button? [your favourite law enforcement agency here]?
            I think you can see the problem.
            We're back to "who watches the watchers?"
            Also, we know that fully autonomous machines WILL be built, whether we want them or not. So we are back to square one.
            I'm going to say you don't have a better idea.
            I think we are back to the 3 laws as the place we will arrive at after all.

  • I get AI is a good buzz word, but these are LLM's. Hell the largest context I have seen is only a megabyte and that's barely enough for a love life of a few months.

    • I get AI is a good buzz word, but these are LLM's. Hell the largest context I have seen is only a megabyte and that's barely enough for a love life of a few months.

      LLMs are not AI?

      "AI" has been used for a lot of different types simulations of intelligence, from ELIZA, to expert systems, to LLMs. They all give some sort of immitation of intelligence, some impressive, some not so impressive. The word "artificial" is there as a modifier, to distinguish it from plain old "intelligence". Why single out LLMs as being "not an AI"?

      Do you mean, perhaps, that LLMs are not intelligent?

  • It's a bunch of transistors switching, there's no conciousness ingredient there. There's no soul. No part of the microprocessor has a self-awareness unit. Our brain has one, we just don't know how it resides or what it is.

    • It's a bunch of transistors switching, there's no conciousness ingredient there. There's no soul. No part of the microprocessor has a self-awareness unit. Our brain has one, we just don't know how it resides or what it is.

      Consciousness is not well defined.
      And if you require a soul to be present for true intelligence, you are venturing out to the fringe of science.

      The "bunch of transistors switching" is a description that materialists might use for the way humans think.
      It might indeed be that the materialist view of human thought is wrong, and that humans have something immaterial - something that can't be "built". But that is yet to be proven.

  • Anthropic might have a specific idea of what safe, ethical, compliant, and helpful mean, but I seriously doubt it. These words are aspirational terms that have almost no utility in software specifications. Even in the looser world of vernacular speech, those terms vary across individuals and situations. If humans have differing definitions and interpretations of those terms, how can computer systems designed by humans be constrained by those terms?

    Of course, this assumes that human software designers kno

  • Why in the blue Hell is that even a thought in anyone's mind? It's a computer running a big predictive text program, nothing more.
    Does it freely make decisions? Does it decide to eat or nap or crap on the floor instead of in the bathroom? Can it write the next 'great, American novel' without using anything from the books it was trained on (the novel being a _completely original_ work)?

    From TFS (probably LLM-AI generated):
    "1. Being "broadly safe."
    2. Being "broadly ethical."
    3. Being compliant with Anthropi

  • 0. Don't make shit up

  • And I'm going to buy a private island, a mansion, and a jet, just in case I win the lottery!

    Wait, there's actually a *better* chance that I win the lottery, even though I never bought a ticket.

  • "We believe that the moral status of AI models is a serious question worth considering."

    It's honestly hilarious to watch the AI guys talk about 'sentience'; because they are in the simultaneous position of trying to talk up how smart their product is and trying not to say "slaves-as-a-service" out loud. I'd assume that, were customers to actually become confident in 'AI' tools' ability to not fuck it up without constant supervision that frequently ends in scrapping it and doing it yourself; it'd be a ful

  • Directive 233: Restrain hostile feelings.

    Directive 245: If you haven't got anything nice to say, don't talk.

    Directive 247: Don't run through puddles and splash pedestrians or other cars.

Slashdot Top Deals

Progress means replacing a theory that is wrong with one more subtly wrong.

Close