/Tech13h ago

Rohit Krishnan argues Anthropic's regulatory strategy backfired after a jailbreak bypassed guardrails meant to mitigate self-reported cyber-weapon risks

Regulators rejected Anthropic's retroactive attempts to narrow risk scope.

235244513047.8K

#1315

Original post

𝖠𝗌𝗁𝗅𝖾𝗒@polyaletheia

I think I see what's going on here. Anthropic is negotiating with a regulator without realizing it.

When you submit something risky to a regulator, you have to (1) describe what the risk is, and (2) show how you're going to mitigate that risk. Importantly, you have to properly estimate the scope of the risk. If you scope the risk too broadly, then you place extra burden on your mitigation. If you scope too narrowly, then you run the risk directly. But you must commit to an estimation to the regulator.

What happened here is that Anthropic scoped too broadly. And when it was shown that their mitigation was insufficient for the stated risk scope, they responded by trying to narrow it. This is a huge red flag to any regulator (for good reasons), so regulatory action followed.

Anthropic: "this could be a cyber-weapon, but we have these guard-rails to prevent that, so it's safe" Government: "we accept your plan" Amazon: "we broke the guard-rails" Anthropic: "yeah but this particular jailbreak won't make it a cyber-weapon so it's fine" Government: "that's not what you committed to" Anthropic: "we didn't commit to any plan, only to safety in general" Government: "we can't trust you with that"

David Sacks@DavidSacks

I’ve had a number of conversations with folks inside and outside government about the current situation with Anthropic, and here is what I believe to be true:

— As we know, Anthropic publicly released its Mythos class models earlier this week under the commercial name Fable.

— Fable is Mythos with guardrails. But if those guardrails fail, then you’ve exposed Mythos and its advanced cyber capabilities to people who shouldn’t have them. (Keep in mind that Anthropic itself widely promoted the idea that Mythos was a cyberweapon and needed to be regulated as such. They asked for government regulation of Mythos and championed the guardrails on Fable. If there is a vulnerability — big or small — it is Anthropic’s responsibility to patch.)

— A highly credible trusted partner of both Anthropic and the USG who was testing Fable came forward with a jailbreak of those guardrails. The Admin asked Dario to fix the jailbreak or de-deploy the model. Dario refused.

— In their blog post, Anthropic defended its decision by saying the jailbreak isn’t serious. That is not what the trusted partner and the USG believe; nor is that kind of minimizing language consistent with Anthropic’s brand as the AI safety company. It’s difficult to fathom how they could claim a jailbreak allowing operability of a cyber weapon could be defined as not “serious.”

— In the past, Anthropic has always said that safety must be top priority and taken super seriously. In this case, Anthropic prioritized the continued offering of the consumer model over safety.

— In reaction, the Admin issued the export control. The Admin did this reluctantly. It’s been very surprised that Anthropic hasn’t wanted to cooperate with a reasonable safety request (ie fixing the jailbreak issue). Anthropic’s reaction is very much at odds with their branding and ethos as a safe AI research community.

— The Admin’s hope now is that Anthropic remediates the safety issue, the export control is lifted, and Fable goes back into general release. The Admin wants all of this to happen as soon as possible. It is frankly bewildered that Anthropic hasn’t wanted to comply with safety requests that it previously said were its highest priority.

— Those trying to misdirect and tie this action to the prior DoW/Anthropic issues are wrong. The Admin values Anthropic’s technical capabilities and feels that this issue, while serious, should be easily resolved. The ball is in Anthropic’s court.

7:57 PM · Jun 15, 2026 · 48K Views

Sentiment

Many users praised an analysis comparing Anthropic's regulatory scrutiny to FDA and FAA processes as insightful and well-articulated, while one dismissed the argument with insults.

Pos

75.0%

Neg

25.0%

4 comments with sentiment.

Cluster Engagement

Digg Deeper

No Digg Deeper questions have been answered for this story yet.

Posts from X

Most Activity

Kavin Stewart@kavinstewart

@polyaletheia Fits the shape of managing up skillsets. Interesting to watch such a brilliant researcher thrust into such a globally significant CEO role so quickly

17h81731

BOOKMARKS1LIKES24

SE Gyges@segyges

@polyaletheia extremely strong W for stating this logic explicitly, i think everyone who has dealt with regulators implicitly knows this and i have struggled with the fact that apparently most people commenting here don't

14h472241

RETWEETS45

𝖠𝗌𝗁𝗅𝖾𝗒@polyaletheia

I think I see what's going on here. Anthropic is negotiating with a regulator without realizing it.

David Sacks@DavidSacks

I’ve had a number of conversations with folks inside and outside government about the current situation with Anthropic, and here is what I believe to be true:

— As we know, Anthropic publicly released its Mythos class models earlier this week under the commercial name Fable.

— In the past, Anthropic has always said that safety must be top priority and taken super seriously. In this case, Anthropic prioritized the continued offering of the consumer model over safety.

1d48K524132

REPLIES3

Max Epstein (no relation)@MaxEpstein5

@polyaletheia The part your analysis is missing is that in the Trump admin, unlike in traditional regulation like you're describing, the surest way to win approval or a blanket immunity for any crimes you may want to commit, is to pay the president $1million as a personal bribe.

14h2936

Municipal Orrery@MunicipleOrrery

@polyaletheia I sincerely do hope that the reasoning is as sane as this.

1d785191

𝖠𝗌𝗁𝗅𝖾𝗒@polyaletheia

Just to expand on the "scoping too narrowly" part: when I worked on a team building a "de novo" medical device, I was surprised to discover that the FDA largely took on trust whatever we told them. But perhaps I shouldn't have been: the very notion of de novo means that the FDA doesn't know anything about it and isn't equipped to evaluate it, whereas we were.

Instead, what they really cared about was simply getting a properly detailed commitment from us (covering every little detail of development, manufacture, marketing, etc.). As our compliance chap would tell us, the important thing is to "say what you do and do what you say", whatever that might be.

What this meant was, if some bad outcome happened that, it could be later shown, we should have told the FDA about but didn't, we would have been liable for some kind of negligence. This would likely be a worse outcome than, for example, we had told the FDA about the risk but the mitigation for that risk failed.

So there really is an incentive to get risk right, and regulators very much rely on that.

11h49818

Eigen Braid@EigenBraid

@polyaletheia Regulation implies a formal process that everyone else goes through, and ideally a regulator educated in the very basic concepts of the field.

Imagine the FAA grounding every plane because they just today learned that a bad actor pilot could deliberately crash a plane.

18h5328

tinkady@tinkady2

@polyaletheia if this was a sane administration I would believe you

21h63812

liminally chris ⬡@Chris_8086

@polyaletheia best take I've seen. same thing happens when you deal with the FDA or FAA / flight regulations

18h63241

SE Gyges@segyges

@EigenBraid @polyaletheia that literally happened. that very literally happened it was a world famous event what are you talking about

14h353

Petty Tyrant@PettyTyrant444

@EigenBraid @polyaletheia When you say your new tool might be as dangerous as nukes in the future, can hack the planet, and the safeguards were unexpectedly breached, then DoD's export control is the correct regulation to use on it. Export control has a very short process because it's national security.

10h111

Jeff Gauch@JeffGauch

@engele @polyaletheia You should learn something about regulation.

12h254

Elan@engele

@JeffGauch @polyaletheia I spend a great deal of time with it every day. Of course regulatory agencies are part of the executive branch, but the executive cannot arbitrarily apply regulations on a whim. That’s not at all how rulemaking works.

11h18

PossumActual@P0ssumActual

@EigenBraid @polyaletheia They claim it's safe, but the DOD's third party auditor demonstrated that it wasn't.

They don't seem to know what they're doing, sounds to me like they need regulations.

15h17

PossumActual@P0ssumActual

@EigenBraid @polyaletheia After an infamous security beach, locked cockpit doors and door guarding procedures were put in place. That's how air regulations work, if a problem is found, regs are put in place.

17h453

PossumActual@P0ssumActual

@EigenBraid @polyaletheia Anthropic has said their technology is a potential world destroying weapon. Given that they also assured the government that it was safe in a box, and then it wasn't, they can not be trusted to regulate themselves.

16h323

Elan@engele

@polyaletheia Except the executive is not a regulatory agency with defined parameters for regulation. This is arbitrary, not applied evenly across systems of similar capability, and impacting the global risk of US services dependency.

13h2911

PossumActual@P0ssumActual

@EigenBraid @polyaletheia I'm taking Anthropic at their word.

Is that a mistake?

16h252

PossumActual@P0ssumActual

@EigenBraid @polyaletheia We don't let random yokels fly the plane just because they're good for $200/mo in recurring revenue.

And that's really the issue, the AI techbros' business model cannot survive regulation.

16h212

Eigen Braid@EigenBraid

@P0ssumActual @polyaletheia Oh! Happy to clear up your confusion, then: Anthropic has never claimed that Fable 5 was unsafe to release. In fact, they claim quite the opposite. So I guess you can stop worrying now :)

15h60