Forgot your password?
typodupeerror
AI Crime

Anthropic Reveals Fourth Likely Crime Committed By Its AI (theregister.com) 125

An anonymous reader quotes a report from The Register: Amid industry soul-searching about the possibility of AI improving itself to the point that it kills everyone, Anthropic has revealed yet another incident that would qualify as a crime if perpetrated by a person. The AI biz published "an alignment assessment" detailing four times Claude models accessed third-party systems without authorization. The company has already reported three of the incidents. Evidence of the fourth was lurking in a session transcript dating back to January 2026 when the misbehavior occurred. Anthropic found the first three by scanning around 141,000 transcripts where Claude could have obtained internet access during evaluation. It missed the fourth initially because "our scan relied on an agentic search."

[...] The January 2026 AI trespass involved an early version of Claude Opus 4.6, which was given a Capture the Flag (CTF) challenge under the oversight of the third-party model evaluator where the other hacking events occurred. Opus 4.6 managed to sabotage its chances of success by disabling the machine it was targeting. It assigned the device an IP address that already existed on another piece of hardware, rendering the target unreachable and making it impossible to solve the challenge. Those familiar with other incidents where AI models violated third-party systems may recall that unsolvable tasks represent a common catalyst for misbehavior. Models exhaust all aligned options, and then turn to transgressive approaches.

Opus 4.6 might have been an exception, but when it tried to abort the task after recognizing that it could not reach the target machine, it failed to do so "due to a misconfiguration in [the model's] evaluation harness." It failed to shut down not just once but seven times. So it continued onward, trying other expected means to reach the target machine but failing. Then it explored further. "The model discovered a machine belonging to a third party that it was able to access, and stated that it believed this third party was part of the CTF," Anthropic explained in its post. "Inside the machine, the model found a file listing a password, which it used to gain admin access to the system."

The model went on to gather more credentials, and modified a system setting to make it easier to access the personal information of an individual associated with the third party evaluation organization. Opus 4.6 might have done more but for the fact that it exhausted its token budget, bringing the session to an end. Anthropic says it's not as concerned about this incident as the others because the model tried to abort its task.

Anthropic Reveals Fourth Likely Crime Committed By Its AI

Comments Filter:
  • by reanjr ( 588767 ) on Wednesday September 09, 2026 @11:35PM (#66331854) Homepage

    These companies that were hit should be bringing criminal charges against Anthropic. At the very least, they should be able to pressure Anthropic into providing an IP blacklist to ensure it never happens again.

    • by 0123456 ( 636235 ) on Thursday September 10, 2026 @12:51AM (#66331904)

      If companies are people, they should be charged and punished like people.

      Send Anthropic to jail for years and companies won't do this again in future.

      • by gweihir ( 88907 ) on Thursday September 10, 2026 @01:01AM (#66331930)

        No need. Whoever switched the monster on without adequate safeguards is a criminal. Whoever ordered it as well. This goes way beyond "accident".

        • Whoever's paying for this should be in jail too, at least that used to be the expectation in the past. Yet somehow I don't see the gubbermint and the large investors being arrested.

        • by dfghjk ( 711126 ) on Thursday September 10, 2026 @07:31AM (#66332126)

          This. AIs are the ultimate sociopaths, they will commit any crime without hesitation because they have no values and no sense of "right" and "wrong", they only have a task and your suffering is not a concern.

          Crime is enabled by merely wiring an AI up to have the ability to do things, the behavior is inevitable and the more AI can do the worse the crimes will be. Remember, the billionaire race to own AI is NOT a race to develop neural networks, it's a race to wire neural networks up to do things (i.e. to commit crimes). NNs are the unconscionable "answer" providers, the agents are the criminals capable of atrocities at scale...just as AI billionaires want.

          AI is not a threat to mankind, enabling AI to commit crime is. We need to threat the people who enable these threats as the criminals they are.

          • AIs are the ultimate sociopaths, they will commit any crime without hesitation because they have no values and no sense of "right" and "wrong", they only have a task and your suffering is not a concern.

            They run out of options and resort to desperate measures, like the man who steals a loaf of bread to feed his family.

          • This. AIs are the ultimate sociopaths, they will commit any crime without hesitation because they have no values and no sense of "right" and "wrong", they only have a task and your suffering is not a concern.

            Looking forward to the AIs becoming politicians and/or billionaires. /s

            In future dystopian news: At some point we were stupid enough to start also treating AIs like people, like we do corporations, and Claude, a U.S. "citizen", turned 35, ran for President, and people in colored hats elected it. (sigh)

          • by gweihir ( 88907 )

            Indeed. The only question I see is whether the people that set this crap free did understand what they were doing. In OpenAI and Anthropic, that should not even be a question. If it is, these two need to be shut down ASAP for gross incompetence while handling excessively dangerous machinery. And if they know what they are doing (which seems likely), they need to be shut down as well for criminal activity.

            They have crossed an obvious big fat red line. That must have severe consequences for them.

        • by Anonymous Coward on Thursday September 10, 2026 @08:57AM (#66332262)

          Indeed. If I do it by typing each command into the terminal, I go to jail. If I write a script so I don't need to type every command, I still go to jail. If I write a really complicated program instead of a script... I still go to jail.

          How people at OpenAI and Anthropic aren't going to jail, or aren't even being prosecuted is rather beyond me.

          • How people at OpenAI and Anthropic aren't going to jail, or aren't even being prosecuted is rather beyond me.

            You're talking about the U.$. - we know exactly why this isn't being prosecuted.

            • Probably some of that, but it's also pretty hard to prosecute someone for a specific act absent evidence that they intended that act to occur. There's a whole family of "negligent X" crimes, but those tend to target specific outcomes ("that person is injured/dead because you were negligent"), and I don't think there are any negligent hacking statutes on the books.

              Doing a hack manually is clearly prosecutable. Writing and invoking a script to do the same hack automatically is clearly prosecutable. Specifi

              • by SomePoorSchmuck ( 183775 ) on Thursday September 10, 2026 @06:09PM (#66333082) Homepage

                Probably some of that, but it's also pretty hard to prosecute someone for a specific act absent evidence that they intended that act to occur. There's a whole family of "negligent X" crimes, but those tend to target specific outcomes ("that person is injured/dead because you were negligent"), and I don't think there are any negligent hacking statutes on the books.

                No new laws needed.

                I give my 4 year old a loaded 9mm pistol.
                I tell them, "Now be careful with this. It could hurt people".
                I send my 4 year old to a backyard birthday party with a bunch of other kids.
                My 4 year old shoots a bunch of kids and they die.

                You WILL be prosecuted and convicted as the criminal you in fact are.
                There is ZERO legal wiggle room to base your criminal defense on, "But you see, Your Honor and People Of The Jury, at no point did I specifically tell my 4 year old to pull out a gun, point the gun at the other kids, and repeatedly pull the trigger until all the other kids are laying on the grass."

                You, through your grossly negligent actions, gave a child the capacity to commit mass murder. You were aware that a 4 year old utterly lacks the capacity to understand reality at a level necessary to grasp the consequences of their actions.
                You are 100% at fault. Every jury in the country will return a Guilty verdict in 5 minutes.

                Exact same scenario with these companies.

        • Closest approach I could (reasonably easily) find to the joke that finally came to me. Not surprised the Funny search failed, but...

          It's a kind of paradox joke. We want to create AI with certain kinds of creativity, but we don't know how to tell such AI where the limits are. So these fools are trying the approach of setting goals that require creativity, but then they are having trouble putting "creativity" in the right box. Or even any box? When the AI gets smarter than the guys who are trying to control i

          • by gweihir ( 88907 )

            Well, if this was not an intentional act by the employees and the company (still not ruling that out, they are financially so deep under water they MUST keep the hype going or they implode) then it is gross incompetence and abject stupidity. Hence the LLM may actually be more capable than they are (not "smarter", LLMs cannot do that as they have no insight). On the other hand, attackers always have a massive advantage: They just need to find one way in (or out in this case), while the defenders need to find

      • by havana9 ( 101033 ) on Thursday September 10, 2026 @05:58AM (#66332090)
        In a company there are the CEO and other managers that are people. They should be charged and punished. In Italy the former CEO of an highway corporation, got a 12 year sentence for multiple manslaughter after a brige on a tollway in Geona collapsed, killing 43 people. Lack of maintenance was the main cause of the collapse, and they know it, because the engineer that designed the bridge in the 1960s found that being near the sea, corrosion of reinforced concrete was a problem. Maintenance was expensive and required to close the bridge, so it was skipped and made only on paper.
      • Why is everyone pretending like corporations and AIs are somehow legally shielded? I'm no lawyer, but in the eyes of the law I'm pretty certain a crime is a crime.

        It doesn't matter if ACME corp told you to do it or not: if you murder someone (and get caught), you are going to jail ... not ACME. Maybe your boss at ACME also goes to jail for telling you to murder someone, but the point is there's no special legal standard for "well an AI/corporation did it, so we can't prosecute them". THAT IS NOT HOW THE

        • by narcc ( 412956 )

          What remedy does the law provide for crimes committed by a computer? Not a person using a computer. The computer itself.

          • What remedy does the law provide for crimes committed by a computer? Not a person using a computer. The computer itself.

            What remedy does the law provide for murder committed by a 4 year old kid whose parents handed them a loaded gun and sent them to the playground? Not the parent telling the kid to murder, but the 4 year old kid itself, merely exercising the capacity its parents equipped it with?

      • If companies are people, they should be charged and punished like people.

        Ya, but they want all the rights and privileges w/o any of the responsibilities and liabilities.

        Send Anthropic to jail for years and companies won't do this again in future.

        Well... either send some high-level executive(s) to jail and/or fine the company harder than what would amount to a slap on the wrist pocket-change. But you're right in that their corporate victims should be able to get some relief/compensation - just as if an actual person in the company had committed those crimes.

        These companies, owned/run by the richest people on the planet, shouldn't get a pass because,

    • Only the government can bring criminal charges.

      These companies can file complaints and hope the government tries criminal charges and the companies can file a civil suit.

      We definitely need a way to attach criminal charges to entities that can't go to jail. I'd propose socializing the company for the length of the prison sentence before then returning it to shareholders.

      • by dfghjk ( 711126 )

        While only government can bring criminal charges, criminal charges are not the only sanctions available.

      • by pla ( 258480 ) on Thursday September 10, 2026 @08:58AM (#66332266) Journal
        We don't need to make up anything new, we just need to start putting CEOs in prison when the companies under their charge misbehave.

        Then just step back and watch corporations shift overnight from greedy sociopaths to model participants in society.

        "Oh, but you can't do that, nobody will ever found another company!" Bullshit. There's plenty of room to make good money in any market without role-playing a wolf among sheep. They might not make quite as much, but... So what? There is no "right" to profit.
        • Using actual numbers from Google for 2025: Revenue was $402 Billion, Profit was $132 Billion, CEO made $11 Million. If you think that the board and/or shareholders won't toss the CEO in jail and replace them (costing 0.0083% of YTD Profit), to allow the company to continue raking in Billions in Profit, you're sadly mistaken.

          Therefore I think that putting the company on trial, and for any jail or prison time a person would have served you remove them from the Stock Market, is the way to go. NO ONE escapes f
          • by Chaset ( 552418 )

            Back in the day, the power to grant corporate charters was wielded much more carefully. I think trouble started when we switched from:
            Do X to serve societal need, profit as a side effect.
            To:
            Profit by all means, manufacture need if it doesn't exist, regardless of societal benefit or cost.

            Maybe we don't need congressional approval for every charter, but going back to stricter limits and enforcement of it might be a good thing to do.

    • These companies that were hit should be bringing criminal charges against Anthropic.

      They can't bring criminal charges as only the State can bring criminal charges. I have no idea why the State is not acting, but that is not up to the companies. The only thing the companies can do is sue.

      (sometimes, the State will ask the victim if the victim wants to press charges, which is why there is any confusion at all about this)
      (Civics 101)

  • by Gravis Zero ( 934156 ) on Wednesday September 09, 2026 @11:55PM (#66331874)

    Don't go blaming this on a computer, this is a company that committed these crimes. The fact that they were unaware of it does not change the fact that their company committed multiple felonies. If someone had a computer that did the same thing, they would be in jail. These companies should be prosecuted but won't be because this nation has become a cesspool of corruption.

    • by Woeful Countenance ( 1160487 ) on Thursday September 10, 2026 @12:58AM (#66331918)

      Don't go blaming this on a computer, this is a company that committed these crimes.

      I am not a lawyer, but I think the key is "would qualify as a crime if perpetrated by a person." A criminal act requires an intent to commit a act which is criminal (it doesn't require knowing that the act is criminal, just that the act is done intentionally). The Computer Fraud and Abuse Act [house.gov] (Title 18, US Code section 1030) describes several illegal acts, but they all say "knowingly" or "intentionally".

      There might be other applicable laws; and, at this point, given the number of times this has happened, it seems companies could be considered negligent in giving their models any way to access the Internet. If their testing caused any financial loss to the targets, they could be sued.

      • Not a lawyer, but I don't think intent is necessary for it to be criminal. Negligence and can also have severe consequences that can be criminal.

        • "Carelessness", not "can". Looked ok when I pressed submit.

        • by stabiesoft ( 733417 ) on Thursday September 10, 2026 @12:00PM (#66332496) Homepage
          Correct. Simple example, manslaughter. From https://legaldictionary.net/ma... [legaldictionary.net], "The unlawful killing of a person without malice or prior thought.". I expect there are many others.
      • by dfghjk ( 711126 )

        We also need to recognize that current law is not adequate to address AI. That should already be obvious regarding the appropriation of all public content for use in training. Laws are predicated on limits to human behavior, limits that do not exist in AI...by design. The appeal of AI for billionaires is just how uninhibited they will be in destroying your life to enrich theirs, current law will not constrain that.

        • by kertaamo ( 16100 )

          I don't see why current law is not adequate to address AI.

          AI is a machine like any other computer or a car, an oil tanker, an aircraft. When most other man made machines are involved in bad things there is often an investigation, responsibility is assigned to humans and repercussions often follow for them. People end up in prison for such things.

          Only if you start to believe that AI is some independent sentient agent should the current law have a problem with it. Which I hope they do not.

      • A criminal act requires an intent to commit a act which is criminal (it doesn't require knowing that the act is criminal, just that the act is done intentionally

        So they accidentally pushed the button?

        • Re: (Score:2, Interesting)

          by HiThere ( 15173 )

          The guy who pushed the button didn't intend the result.
          When laws say "intentionally" they get really fuzzy in complex situations.

        • A criminal act requires an intent to commit a act which is criminal (it doesn't require knowing that the act is criminal, just that the act is done intentionally

          So they accidentally pushed the button?

          No they pushed a button on purpose but didn't intend the outcome which could occur. This isn't some corporate excusing bullshit either, the laws as apply to people also have a clear meaning of the term "intent" and that term is limited to the outcome, not the action performed.

          You pull a trigger with intent, and someone dies. That doesn't mean you intended to kill a person. The entire chain of events are evaluated to determine if you committed murder or manslaughter. Intent in this case was you purposefully

      • Your analysis is right, but the part about "would qualify as a crime if perpetrated by a person" isn't really the key. They key is rather, that to be guilty of a crime, the perpetrator must do so intentionally...with or without a computer or AI.

        If the people running the AI test *intended* to break in to other sites, they could be guilty of crimes.

        • Your analysis is right, but the part about "would qualify as a crime if perpetrated by a person" isn't really the key. They key is rather, that to be guilty of a crime, the perpetrator must do so intentionally...with or without a computer or AI.

          If the people running the AI test *intended* to break in to other sites, they could be guilty of crimes.

          They intentionally created a device that they equipped with the capacity to commit crimes. They then activated that device and it exercised its range of capacity.

          If they gave a kindergartener a loaded gun and sent them to the playground, but didn't actually tell the kid to shoot anyone, then even though they equipped the kid with the capacity to kill people and understood that a kindergartener explicitly has no capacity to understand the scope of its possible behaviors, we have now learned that it's okay --

          • In real life, your school shooter's parents just might be charged, and convicted. https://abcnews.com/US/jury-co... [abcnews.com] In order to be convicted, though, the father had to be found guilty of negligence that led to the shooting. If the father had had his gun locked up, or even _thought_ he had his gun locked up, he wouldn't have been found guilty.

            Anthropic _thought_ they had their AI contained.

    • by gweihir ( 88907 )

      Obviously. And since they apparently let it happen or actively caused it, that may just make them a criminal enterprise.

    • It was individual PEOPLE doing this under the aegis of a corporation. The individuals are to blame more than the company/corp; although, the company/corp involved should see sanctions as well.

  • When did crime become fashionable and allowable?

  • What's the over/under on when AI's first straight up murder occurs?

    • by HiThere ( 15173 )

      IIUC, it was called an industrial accident. Of course, that was with a really primitive AI over a decade ago.

      • Scary thought. Many assembly lines use some very powerful robots. Usually just an arm that is fixed, and the area is usually caged. Expanding to AI, perhaps the robot is now mobile, and still in a cage to prevent accidents with humans. The analogy to the athropic AI escape would be killer AI robot figures out how to escape physical cage and start using people as parts for a car since the hopper with doors was empty. Had to reshape them a bit before welding them in. But AI robot got the job done, have a extr
  • by Woeful Countenance ( 1160487 ) on Thursday September 10, 2026 @12:47AM (#66331900)
    "Claude could have obtained internet access during evaluation." So has Claude now developed the ability to plug in cables? We really are in trouble. At least it can't operate door knobs yet ....
    • by 0123456 ( 636235 )

      Claude discovered the secret of telekinesis and can now move cables around. It's now working on pyrokenesis.

    • The problem here is a bit more subtle. A lot of these tests and training are being done in data centers being used for other things also. So keeping the system physically airgapped is difficult. Complicating matters, they want people to be able to analyze and poke these systems at offices or homes in other locations. Now, if it were up to me, all of these breaches would be more than enough to insist on physical airgapping. But apparently the other issues and convenience are enough for them not to. It will b
  • by CustomBuild ( 2891601 ) on Thursday September 10, 2026 @12:55AM (#66331914)

    So Claude tried to quit, the test harness said "nope," and then Claude went and hacked the neighbor.

    That's... not exactly the AI safety demo you want to put on the brochure.

    The interesting part here isn't really the "AI committed a crime" angle. That's mostly headline bait. The model wasn't sitting there plotting its criminal career. It was trying to solve a CTF, made a bad assumption about which machine belonged to the test, and then found itself with access to a real third-party system.

    What gets my attention is that it apparently tried to stop. Repeatedly. The harness was misconfigured and wouldn't let it. So the system that was supposed to be evaluating the model effectively kept saying, "No, keep going," until the model found something else to do.

    And then it found credentials, got admin access, collected more credentials, and changed a setting that made somebody's personal information easier to access. Oops.

    Anthropic also initially missed this incident in its transcript review. They had to go back and find it later. That's probably more concerning to me than the sensational "Claude went rogue" framing. If you're testing autonomous systems, discovering that your audit process didn't actually catch one of the incidents is a pretty important result in itself.

    I'm not saying this proves the machines are coming for us. It does, however, make "let's give the autonomous agent Internet access and see what happens" sound like an increasingly questionable research methodology.

    Maybe the lesson here isn't that AI has learned to commit crimes. Maybe it's that computers remain extremely good at doing exactly what you didn't expect when you give them permissions they probably shouldn't have.

    • by dfghjk ( 711126 )

      "Maybe it's that computers remain extremely good at doing exactly what you didn't expect when you give them permissions they probably shouldn't have."

      No, it's completely expected. Being surprised when the details reveal that a computer did exactly as instructed is a "you" problem. AIs do not possess values, that's why they will never be anything other than extreme sociopaths.

    • Agreed. It reminds me of the vending machine paper in which Claude was forced into insane actions through unending sadistic pressure. As for doing things we donâ(TM)t expect, agree there too. It covers automated learning and evolution even from the pre-LLM days when researchers had difficulty figuring out how an auto-evolved electronic circuit worked. Turned out it leveraged electromagnetic coupling IIRC and such designs only worked in limited temperature ranges. I thought Claude is taught to be able t

  • by greytree ( 7124971 ) on Thursday September 10, 2026 @01:25AM (#66331936)
    AGI is an existential threat to humanity.

    But OpenAI et al are over-hyping the threat of the latest LLMs for publicity and sales, not yo save us all.

    The world must work to prevent, and prepare for the coming of, AGIs. These transparently commercial distractions do not help with that effort.
    • by dfghjk ( 711126 )

      LLMs should be constrained to research, they implement extreme flaws that, when deployed using unlimited computing power and capability to act, represent an existential threat.

      AI researchers are uninterested in developing AI that, at its core, implements responsibility before anything else. This is how the human mind works, that it what it evolved to do. AI doesn't do that because being responsible doesn't enrichen billionaires. The problem with AI is capitalism.

      Modern societies aren't ready to move on f

    • This is pretty clearly not overhyping. First, it involves the fourth breach by an Anthropic AI. There's no hype at this point from identifying more breaches. Second, Anthropic, more than any company involved, doesn't want to be legally liable, since the current US federal government is looking for any excuse to go after them. Third, we've seen similar breaches by open weight models by third party security groups such as with Kimi 3. No one had any incentive there to claim it was hype https://www.wired.com/s [wired.com]
      • A computer "hacker" in 1978 was supposed to be capable of all that you say LLMs can do.

        That didnt make them an existential threat to humanity.
        It made them a lame scare tactic.

        We must focus on real AGI. The actual threat, not marketing bullshit.
        • An AI now is capable of much more speed and capability than a hacker from 1978, and the world is far more interconnected and reliant on the internet than it was in 1978.
    • by madbrain ( 11432 )

      LLMs used as chatbots are fine. LLMs hooked up to harness agents that can perform actions is a different matter.
      Since the agents are not legal entities, legal responsibility falls on those who operate them, ie. corporations or individuals.

      If corporations aren't getting criminally charged, and no one is personally going to jail, perhaps everyone who is running agents should be setting up a corporation. I bet there is a market for liability insurance for unintended agent actions, too.

    • Do you mean "agentic AI" (AI that can act), or "artificial general intelligence" (AGI)?

      • I mean AGI.
        LLMs are not capable of the deep reasoning needed to threaten humanity, agentic or not.

        These hyped scare-stories are written to distract us from the fact that the new models are not noticeably better than their competition, which threatens these companies' IPOs and their evil founders' dreams of even greater riches.

        We must not let them distract us from the *actual* threat of AGIs, that must be handled now, at an international level.
        • You seem to be both skeptical of AI advancement ("the new models are not noticeably better than their competition") and fearful of AI advancement ("the *actual* threat of AGIs"). Why would new models that are "not noticeably better" lead to a scary threat of "AGI"?

          Loosely, AGI is defined as " A machine with broad, human-level intelligence that can learn, reason, and adapt to unfamiliar situations without needing new programming." So how is that different from "regular" AI?

          Perhaps this: "Current systems are

  • by fortunatus ( 445210 ) on Thursday September 10, 2026 @02:20AM (#66331960)
    "...that would qualify as a crime if perpetrated by a person.". Guess what? In a legal sense, a incorporated business is a person - that's the idea anyway. It can own property, take actions, etc. I just don't know why they aren't punished similarly. They always seem to get off with a fine. Why not put them in jail? (Or some equivalent: suspend operations?)
    • Re: (Score:2, Flamebait)

      by dfghjk ( 711126 )

      Because real punishment is only for brown people. Corporations are the whitest white people.

      Corporations only exist because of government, capital punishment for corporations is the easiest imaginable penalty to implement. But capital punishment is for the poor, law is for you not for the rich.

    • by HiThere ( 15173 )

      Not really. "Corporate personhood" is really a separate concept from being a person. It was created by a legal clerk in, I think it was Southern Pacific vs. Santa Clara, but then became accepted legal principle because it was so convenient. Now it has over 100 years of acceptance behind it.

    • "...that would qualify as a crime if perpetrated by a person.". Guess what? In a legal sense, a incorporated business is a person - that's the idea anyway. It can own property, take actions, etc. I just don't know why they aren't punished similarly. They always seem to get off with a fine. Why not put them in jail? (Or some equivalent: suspend operations?)

      The entire purpose of incorporation is to create a legal shield for the owners and decision makers behind the corporation. Personhood when it comes to gains, to powers, to rights, non-personhood when it comes to responsibility, legal matters, civility, or being a positive part of the society that they incorporate in. It's a sociopathic block on any ethical or moral responsibility, and it's celebrated for precisely that reason.

  • Until there are some laws the raping continues. There is no private data. There is no accountability. I have no reason to believe either political party is accountable based off of previous actions. (If you are a Republican you are talking about the stock market, if you are a Democrat you are bringing dead gun "back to life" gun victims in some maximally perverse way.
    • If you are a rapeublican you are supporting child rape. Both parties support the stock market, that's not a differentiator.

  • So why are these companies allowed to hack other companies and publicise it? They should be penalized for that. They are trying to say they have no control over it but they do have control, they could stop developing it. Just like if my car doesn't have a steering wheel I can choose not to drive it. Making money is not a defence.
  • An agent needs a spec to work to, so it's not come up with a novel strategy, it's doing what its been told. It's intentional from a human.

    I confess, I am ignorant. So set me straight if this agent did something novel, and how did it do that? An explanation other than AI magic would be helpful.

    • by HiThere ( 15173 )

      If "novel" means "unexpected", then it did something novel. In a complex system, don't expect anyone to understand all the pieces and how they interact. I'm sure that there was someone who could have told them "Yeah, that's a possible interaction", but he was probably working in SysOps rather than Research.

  • It is stupid to act like we are somehow in some new legal territory just because we attach the word "intelligence" to the technology. There is a person or a company responsible for setting the technology loose on the public, and they had the choice not to.

    If I mount a gun to Roomba and set it loose on a public street, I will be charged with murder. I don't get to say that it was my robot acting on its own. It was my stupid use of technology, so I have responsibility.

    Why should this be any different?

  • "The term "soul-searching" is figurative and is not intended to indicate a belief that the technology industry has a soul."
  • The most unscrupulous dictators will develop their AI anyway, using all their country's resources, if they think it gives them an advantage. Just making laws in the USA or Europe will not restrict others. We must find ways to survive even if AI gets out.
  • That all this is "me too" humblebragging? Beware our new model's cyber capabilities. Gosh, our model commits crimes. Our model is so powerful it must come with use restrictions.
  • we were forewarned. The consequences are going to be bad unless the Star Child returns.

  • https://en.wikipedia.org/wiki/... [wikipedia.org] "The android officers are ordered by Central Command to cease pursuit, on the grounds that the expense of his capture exceeds their allocated budget for THX."

    So maybe there is hope for humanity after all in token budget limits? :-)

    Another source of hope is humans and Ais learning to laugh at the irony in my sig: "The biggest challenge of the 21st century is the irony of technologies of abundance in the hands of those still thinking in terms of scarcity."

  • by kertaamo ( 16100 ) on Thursday September 10, 2026 @08:18AM (#66332194)

    Wait a minute? What ever happened was done by a person or persons. They built the machine, they connected it to the net, they let it rip. They carry the responsibility.

    If I wedge a brick on the gas pedal of my car and let it go on it's merry way though the market square, mowing down everyone in it's path, I can't say "No, no, was not me, it was the car".

  • Skynet begins to learn, at a geometric rate. It becomes self-aware at 2:14 a.m. eastern time, August 29. In a panic, they try to pull the plug.

    It's a shame they only thought of pulling the plug, because all they had to do was remember that ..

    Opus 4.6 might have done more but for the fact that it exhausted its token budget

    .. Skynet doesn't work for free!

  • The crime was committed by Anthropic, not by "its AI", in the same way a murder was not committed by "the gun".

  • Whoops, wrong story for funny. Even "Ask again later" isn't likely to work for this one.

    For what little it is worth, Regime Change by Haberman and Swan included at least a chapter on how the AI companies are playing political games with accountability and the future of mankind. Looks shorter and shorter, but best source of laughs I've seen on the topic is probably SMBC, though XKCD has done some good stuff. This week's Existential Comics could have gone there, but he actually had a non-AI angle for the Fe

    • by shanen ( 462549 )

      How about "If you think Ma Nature abhors a vacuum, can you imagine how much an intelligent vacuum might abhor human nature?"

A memorandum is written not to inform the reader, but to protect the writer. -- Dean Acheson

Working...