Policy scholar Dean W. Ball argues Anthropic's safety policies constitute anticompetitive behavior disguised as AI safety

VIEWS127.3KBOOKMARKS94LIKES619REPLIES22

spor@sporadica

at this point if you're a SWE, you either accept that you won't get better than opus 4.8 from anthropic or you just move to another model/platform entirely

bc why would anyone in their right mind use Fable? esp with how many insane false negatives we've seen on cyber+bio tasks

mattparlmer 🪐 🌷@mattparlmer

Cosign all of this

Speaking as a software engineer, there now a lot of reasons to think that Claude cannot be trusted on difficult performance-critical tasks, going forward I would be irresponsible not to have Codex taking point on anything that touches a GPU

Massive own goal

2d127.3K61994

RETWEETS104

Dean W. Ball@deanwball

My last observation re: Anthropic’s secret sabotage safety policy, is that it undermines actually good safety policy. How?

1. First, it is very plausible to describe this as anti-competitive behavior (even if you are maximally sympathetic to Anthropic here you must admit this), and it is behavior being justified in the name of AI safety. If you believe, as I and many Anthropic staff do, that it may end up being critically important to relax antitrust enforcement so that the frontier labs can cooperate and collaborate on some areas of AI safety, Anthropic just undermined the case for that in a large way.

2. Overall, this massively and profoundly raises the status of the argument that AI safety has been hype to justify monopolistic behavior by labs. I continue to believe AI safety is a real and serious issue that is growing in importance rather than diminishing. If you agree with me, this incident is a setback, maybe a serious one.

3. As I have observed elsewhere, Anthropic’s official corporate policy is structurally identical to the fact pattern alleged against them by the Department of War. I still think DoW acted both falsely and wrongly in that fight, but it is no longer possible to defend Anthropic with a full throat after this incident.

4. This raises the case for heavier handed regulations. Anthropic is making an awfully good case here that their products ought to be treated as utilities, and thus that their alignment practices should be a matter of public policy rather than private property. I am starkly opposed to this sort of state power grab, but Anthropic is doing more to justify it than anyone else.

5. Thus, significant damage has been done to a community and entire approach to AI governance. It was done unilaterally by Anthropic, likely motivated largely by self-interest and justified within the internal psychology of the firm through the lens of safety.

I suspect this is fixable in the economic and legal senses for Anthropic, but I fear the trust that has just been broken, and the goodwill extinguished, will take very much time to repair.

3d142.1K1.1K337

mattparlmer 🪐 🌷@mattparlmer

Cosign all of this

Speaking as a software engineer, there now a lot of reasons to think that Claude cannot be trusted on difficult performance-critical tasks, going forward I would be irresponsible not to have Codex taking point on anything that touches a GPU

Massive own goal

Dean W. Ball@deanwball

My last observation re: Anthropic’s secret sabotage safety policy, is that it undermines actually good safety policy. How?

1. First, it is very plausible to describe this as anti-competitive behavior (even if you are maximally sympathetic to Anthropic here you must admit this), and it is behavior being justified in the name of AI safety. If you believe, as I and many Anthropic staff do, that it may end up being critically important to relax antitrust enforcement so that the frontier labs can cooperate and collaborate on some areas of AI safety, Anthropic just undermined the case for that in a large way.

2. Overall, this massively and profoundly raises the status of the argument that AI safety has been hype to justify monopolistic behavior by labs. I continue to believe AI safety is a real and serious issue that is growing in importance rather than diminishing. If you agree with me, this incident is a setback, maybe a serious one.

3. As I have observed elsewhere, Anthropic’s official corporate policy is structurally identical to the fact pattern alleged against them by the Department of War. I still think DoW acted both falsely and wrongly in that fight, but it is no longer possible to defend Anthropic with a full throat after this incident.

4. This raises the case for heavier handed regulations. Anthropic is making an awfully good case here that their products ought to be treated as utilities, and thus that their alignment practices should be a matter of public policy rather than private property. I am starkly opposed to this sort of state power grab, but Anthropic is doing more to justify it than anyone else.

5. Thus, significant damage has been done to a community and entire approach to AI governance. It was done unilaterally by Anthropic, likely motivated largely by self-interest and justified within the internal psychology of the firm through the lens of safety.

I suspect this is fixable in the economic and legal senses for Anthropic, but I fear the trust that has just been broken, and the goodwill extinguished, will take very much time to repair.

2d77.4K57867

Ben Thompson@benthompson

@deanwball Maybe folks who pushed back on Anthropic's positioning in the Department of War debate actually foresaw *exactly* this type of behavior?

Dean W. Ball@deanwball

@FifthRocket @Noahpinion @benthompson Ben doesn’t even really believe it consistently. He’s more using the amodei position as rhetorical scaffolding to justify his conclusion than he is taking it seriously in my view

2d18.8K12613

Ravid Shwartz Ziv@ziv_ravid

Anthropic's recent moves amounted to spectacular reputational self-destruction in the AI research community, which is too bad, because this community was one of the first to give them credit and use their coding agents. In general, anti-competitive moves are bad, but couching them in safety makes it worse. Anyway, just noting that I called out this entanglement of anti-competition, safety, and self-regulation a long time ago!

2d4.9K906

Boaz Barak@boazbaraktcs

Good take by Dean. While trying to block competition is generally bad, couching it in safety makes it far worse.

Dean W. Ball@deanwball

My last observation re: Anthropic’s secret sabotage safety policy, is that it undermines actually good safety policy. How?

1. First, it is very plausible to describe this as anti-competitive behavior (even if you are maximally sympathetic to Anthropic here you must admit this), and it is behavior being justified in the name of AI safety. If you believe, as I and many Anthropic staff do, that it may end up being critically important to relax antitrust enforcement so that the frontier labs can cooperate and collaborate on some areas of AI safety, Anthropic just undermined the case for that in a large way.

2. Overall, this massively and profoundly raises the status of the argument that AI safety has been hype to justify monopolistic behavior by labs. I continue to believe AI safety is a real and serious issue that is growing in importance rather than diminishing. If you agree with me, this incident is a setback, maybe a serious one.

3. As I have observed elsewhere, Anthropic’s official corporate policy is structurally identical to the fact pattern alleged against them by the Department of War. I still think DoW acted both falsely and wrongly in that fight, but it is no longer possible to defend Anthropic with a full throat after this incident.

4. This raises the case for heavier handed regulations. Anthropic is making an awfully good case here that their products ought to be treated as utilities, and thus that their alignment practices should be a matter of public policy rather than private property. I am starkly opposed to this sort of state power grab, but Anthropic is doing more to justify it than anyone else.

5. Thus, significant damage has been done to a community and entire approach to AI governance. It was done unilaterally by Anthropic, likely motivated largely by self-interest and justified within the internal psychology of the firm through the lens of safety.

I suspect this is fixable in the economic and legal senses for Anthropic, but I fear the trust that has just been broken, and the goodwill extinguished, will take very much time to repair.

2d7.2K737

John David Pressman@jd_pressman

@deanwball I agree with all this but one point: I think saying the trust is there to be repaired is sort of mistaking the causality of why people are so mad. Everyone savvy (including the DoD) already understood that Anthropic's ideology meant there was a risk they supply chain attack you.

3d713234

Andreas Kirsch 🇺🇦@BlackHC

1. How is this anti-competitive behavior?

It is absolutely within their rights to not make their strongest capabilities available to competitors as part of their product offering

This is parallel to the previous scenario with the government. The public has no right or expectation to the strongest coding abilities from their model

2. The AI safety concern is valid. You don't explain your reasoning. RSI is a risk and people believe that we will see RSI within a year. Mythos might be strong enough to kick this off. Avoiding external RSI loops via Mythos is prudent

Dean W. Ball@deanwball

My last observation re: Anthropic’s secret sabotage safety policy, is that it undermines actually good safety policy. How?

1. First, it is very plausible to describe this as anti-competitive behavior (even if you are maximally sympathetic to Anthropic here you must admit this), and it is behavior being justified in the name of AI safety. If you believe, as I and many Anthropic staff do, that it may end up being critically important to relax antitrust enforcement so that the frontier labs can cooperate and collaborate on some areas of AI safety, Anthropic just undermined the case for that in a large way.

2. Overall, this massively and profoundly raises the status of the argument that AI safety has been hype to justify monopolistic behavior by labs. I continue to believe AI safety is a real and serious issue that is growing in importance rather than diminishing. If you agree with me, this incident is a setback, maybe a serious one.

3. As I have observed elsewhere, Anthropic’s official corporate policy is structurally identical to the fact pattern alleged against them by the Department of War. I still think DoW acted both falsely and wrongly in that fight, but it is no longer possible to defend Anthropic with a full throat after this incident.

4. This raises the case for heavier handed regulations. Anthropic is making an awfully good case here that their products ought to be treated as utilities, and thus that their alignment practices should be a matter of public policy rather than private property. I am starkly opposed to this sort of state power grab, but Anthropic is doing more to justify it than anyone else.

5. Thus, significant damage has been done to a community and entire approach to AI governance. It was done unilaterally by Anthropic, likely motivated largely by self-interest and justified within the internal psychology of the firm through the lens of safety.

I suspect this is fixable in the economic and legal senses for Anthropic, but I fear the trust that has just been broken, and the goodwill extinguished, will take very much time to repair.

2d1.7K192

Dean W. Ball@deanwball

I think it would be incredibly difficult to describe the DoW’s behavior as “savvy” here. And I think the anti-Anthropic/anti-safety coalition has been anything but savvy for the last year at least. Nonetheless, Anthropic has given that group a major win, and this comes after a solid 9 months of losses for the “we are anti policy but mostly anthropic and also we have no ideas for what to do about anything” community of AI discourse

3d590162

Ben Thompson@benthompson

You did concede the point in the post I replied to, which is why I replied to it. I do tend to think that affording people one disagrees with more grace at the time of disagreement is probably prudent. To that end, setting aside pedantic points about whatever the original tweet was referring to, I'm not at all shocked by Anthropic's decisions this week, because my analysis at the time was not based on some sort of personal animus or whatever was ascribed to me, but simply incentives and long-run trends. That analysis holds in terms of this decision. People's declarations and promises change, *especially* when they hold themselves to be moral actors. Suggesting that anyone pointing that out is faking their beliefs to make a point risks missing the more important point completely.

2d912112

Dean W. Ball@deanwball

hence why I conceded that exact point! nonetheless, the government was lying when they claimed Anthropic made these threats, as attested by the fact that they don’t make those claims under oath. A suspicion does not justify the policy action the government took. and either way, the point you’re quoting has little to do with this and was instead about the “they say it’s a nuclear weapon so they should expect it to be regulated like a nuclear weapon” line of argument, which I continue to think is fairly thin intellectual gruel. But yes, you were right to mistrust anthropic.

2d1.3K132

Charlie Bullock@CharlieBull0ck

What’s the argument for this being obviously “anti-competitive” (I assume you mean in an antitrust law sense?)

If they were coordinating with other labs, then I would see the argument. But I’ve never seen it argued that it’s unlawful for a company to unilaterally decide to neuter its own product for PR or liability exposure reasons. It might be a bad business decision, but it’s not illegal.

2d56381

Dean W. Ball@deanwball

I did not say it is obviously anti-competitive in the legal sense, I said it is obviously describable (a lawyer would say colorable) as anti-competitive, in both a legal sense, but much more importantly, in a broader sense. It’s clearly anti-competitive for an ai research company to silently sabotage people who try to do ai research using their products. Banning that use case is fine, but the silence and ambiguity of the sabotage is the salient thing.

2d36881

spor@sporadica

a lot of people i've talked to were still using opus 4.6 anyway, so maybe the fallout of this is limited but like, they've kinda destroyed a lot of trust in basically any model they release going fwd and that's hard to get back, i would think

2d45210

John David Pressman@jd_pressman

@deanwball The problem is that this was a mere suspicion, a paranoia. It was nebulous and hard to prove, so the people who had this grievance couldn't really prosecute it. It would be like worrying about Mormon AI supply chain attacking you, valid but hard to prove.

3d7331

Dean W. Ball@deanwball

@AlmostMedia @theojaffee I am too.

2d2624

Hurrairah@hurrim38

@sporadica a frontier model is now a production dependency like any other, and almost nobody has a tested migration path. Eval suites that run against two providers, prompts that aren’t load-bearing on one model’s quirks. The teams treating this as a fire drill are getting a free one.

2d14921

Pierre-Luc@dallairedemers

@deanwball "AI safety" is woke virtue signalling applied to a new object. It's not real, it's a grift. Extracting non-slop content is the only really difficult problem with this tech.

2d14421

Dean W. Ball@deanwball

Ah, I see what you mean. Yes, I do agree there. I think the big issue was they tried to claim, in the court of public opinion, that anthropic literally threatened to do the supply chain attack based on ideology, but in fact Anthropic never literally did that, it was all just suspicion. So the government’s case in court was way weaker, because they had no actual facts. Just suspicion. And the judges, while fairly deferential to the executive on these issues, were scratching their heads about whether such a severe action could be justified on the basis of suspicion alone. The answer now is much likelier to be yes. If Anthropic loses in their DC Circuit panel and appeals to the full court, the government’s attorneys absolutely will bring this up and it may prove to be the straw that breaks the camel’s back.

2d1256

Dean W. Ball@deanwball

@SuperbBias @CharlieBull0ck @theojaffee that is not their publicly stated intent, actually, no. The opposite of that is their publicly stated intent. The literal opposite.

2d1072

Policy scholar Dean W. Ball argues Anthropic's safety policies constitute anticompetitive behavior disguised as AI safety

Story Overview

Safety rhetoric now invites monopoly scrutiny

Trust erosion in governance circles stays unquantified