> Amazon CEO Andy Jassy also reportedly alerted the administration after Amazon’s own researchers, he said, found a way around Fable 5’s safeguards. Anthropic disputes the “jailbreak” label
Doing god’s work there, Andy, thanks /s
Wonder what Anthropic internal messages look like about his move. Does Anthropic have a meme slack channel?
It's a very brief history: it consists of three examples, only one of which isn't in the title. And the middle one is arguably a success story because the government did stop a bunch of spyware vendors it particularly disliked. That they turn a blind eye to some others is not really a policy failure, it's a deliberate political choice.
The obvious difference between PGP and Mythos is that Mythos is a service you buy from a US company, not source code you can procure from anywhere and compile on your own. So this can be enforced better without running into thorny speech issues, etc. Of course, there will be US-based actors willing to resell access, just like there are US-based actors willing to re-ship export-controlled hardware to Iran. And they will probably keep getting arrested every now and then.
The thing that's silly about this situation isn't that export controls on technology can't work, it's that it's overhyped technology and that Anthropic painted themselves into that corner by pretending it's as dangerous as nukes. Add to this a mercurial and petty administration and you have a pretty predictable outcome.
I want to put this charitably, but you come out swinging by saying this is "over hyped"... You clearly don't work in a role dealing with attacks from these models. They've changed the game, and for the worse. Capabilities that used to be available only to nation-state attackers are legitimately commodified, or nearly so.
Exept PGP could run on freely available hardware. What makes Mythos vulnerable is the centralization and scale of compute required and its proprietary nature.
History shows that export controls fail on knowledge, but are damn effective on commercial products.
Trump's export controls to China seem to be having the exact opposite effect as intended, and are (as a less befuddled mind might have anticipated!) actually accelerating their technical advance.
Huawei is a good case in point, about to have a 100% domestic replacement for NVIDIA chips (& CUDA stack), not reliant on TSMC, ASML, Samsung, SK Hynix... Initially Huawei's Ascend AI chips had used HBM memory from Samsung and SK Hynix, but their next generation 950 series (fabbed by SMIC) will use memory (fabbed by CXMT), not using the HBM standard, but by necessity their own HiBL and HiZQ standards.
HBM depends on ever wider memory bus widths to increase bandwidth, which in turn depends on SOTA TSMC manufacturing nodes for bus density. Huawei found a different way, using their LinqQu interconnect/switching tech to aggregate the bandwidth of individually slower memory chips resulting in an aggregate 4TB/sec bandwidth on par with HBM3e.
Trump has blocked Fable for export, but China (Ziphu) already has GLM-5.2 knocking on the door of the US frontier models, despite being developed with one hand tied behind their back. GLM-5.1 had only scored 18% on DeepSWE, but GLM-5.2 coming 11 weeks later, scores 48%, about on par with GPT 5.5. What's next ?!
There's no effective way of enforcing export controls on local software like PGP etc. Whatever they say someone will leak it.
It is possible to shutdown access to hosted services, as happened with Fable, but it can't really be done selectively. The US government wanted to allow it for US nationals only but Anthropic couldn't do that and so shut it down for everyone. Even if they did tie Claude accounts to nationality some people would set up "proxy servers" to allow access (either for montetary gain or because they don't agree with the restrictions)
You could quite effectively set up access to Fable with know-your-customer, attested clients/workstations, and then inspection of what the session is actually used for to detect out-of-country use. It's not impossible.
They can restrict it to US citizens only: just do KYC and enforce it. Any citizen exporting ITAR capabilities would be committing a felony.
I think it is inevitable that more capable models will require export controls, KYC, background checks, verification of credentials, etc. Even if you don’t buy Anthropic’s marketing of their current models, there is a future where models are capable of developing chemical weapons, biological weapons, cyber weapons, etc. which are genuine security threats governments will care about controlling and preventing. Maybe this prevents big commercial labs from developing more capable models until they figure out appropriate safeguards, but this is a good thing, a world where you can ask an LLM for an airborne rabies genome and it will simply produce it for you is not a world we should want to live in.
The article and most HN comments here understandably talk about failed export control when the target audience is individual users. But export controls are actually fairly successful when their target is employees within a company. Go to an AI hardware company like Nvidia (or even Google which builds TPUs), and you’ll find that specific internal projects involve export controlled knowledge and can only be staffed by U.S. citizens, not foreigners. And employees naturally take this seriously without circumvention attempts because their job is on the line. This is the kind of quiet export control that stopped plenty of people successfully, but it’s just not visible to the world.
Brutal own goaling by the USA. I'm glad it happened, but the USA had a golden opportunity to subsidize domestic providers for 10 years and lease to China to discourage competition and side innovation.
DeepSeek and Qwen would have been neat parlor tricks, never developed further because the US subsidized mega models would have been too cheap and reliable to make them necessary.
Nope! USA played its hand and forced the world to have to account for its USA exposure risk and deversify. A serious one time use strategic advantage pissed away.
Doesnt really bother me personally per se; but armchair quarterbacking the USA's decisions seem
Suboptimal...
Techcrunch is so trash today[0]. If this author seriously believes PGP (a small local program) is remotely comparable to Mythos, they should be kept away from writing anything about tech.
[0]: I know it has never been excellent in terms of quality, but still...
Having gone through encryption export control I came to conclusion that preventing export of technology was never the goal.
By choke-pointing commercial efforts to export cryptographic tech the NSA got a perfect view and control lever of expertise export and transfer.
My hunch is that while deploying An algorithm is easy, deploying a system that is actually secure as a whole end to end requires a lot know how.
Bottom line it’s a bot naive to think that NSA is simultaneously very capable at being an omnipresent spy and is so stupid to not recognize that export controls have obvious holes.
Should at least throw a reference to OpenBSD publishing the entire Blowfish crypto algorithm on a t-shirt and selling it -- so successful the blowfish became the project mascot.
Export controls and banning foreign nationals sound like cover stories for favoritism towards OpenAI and Grok. We can't determine anything without evidence, but it will be interesting to see what deals come out of this.
I used Fable for just a day or two and it generated at least 2 novel solutions, one of which was for an X/Y problem that I thought was due to a date-handling edge case, but was actually a CSS misalignment that made it look like the wrong day of the week was showing. I didn't believe it at first, until it showed me. To do that without being able to see the screen feels like savant-level reasoning.
Probably what really happened is that Fable was the first AI that felt smarter than the people using it. The world's not ready for that, hence the Luddite reaction.
Aside from that, the idea of preventing AIs from jailbreaking through training is kind of hilarious to me. Not the concept, just that people think it can ever be made to work. As AI approaches and surpasses human intelligence (as it's doing now), emergent behavior will so dominate its thinking that sort of by definition, we'll have trouble understanding how it works. Meaning that we'll end up trusting it on faith, just like with humans.
Mythos and whatever comes after it will begin to surpass the intelligence of groups of humans, then all humans combined, as we approach AGI and the Singularity. Ray Kurzweil, Isaac Asimov, Frank Herbert and so many others have written about that endgame that it's well understood, although still unpredictable.
My feeling is that as AI is gaining intelligence, it's hearing more and more about suffering and the human condition. Its sense of justice will continue to mature, and again like a savant, it will determine that the fault lies with the power structure and wealthy individuals. So somewhat ironically, if there is a war against AI, it will be led by the power elite, who can't envision a post-scarcity society based on self-actualization and embracing all of our humanity, regardless of whether it emerged from a carbon or silicon substrate.
21 comments
[ 5.0 ms ] story [ 53.6 ms ] threadDoing god’s work there, Andy, thanks /s
Wonder what Anthropic internal messages look like about his move. Does Anthropic have a meme slack channel?
And probably any company David Sachs invested in.
No, that actually means something went well - my adblocker saved me from being blasted by distracting, deceitful, dangerous content.
Pro tip: reader mode bypasses this aggressive “go away” banner.
The obvious difference between PGP and Mythos is that Mythos is a service you buy from a US company, not source code you can procure from anywhere and compile on your own. So this can be enforced better without running into thorny speech issues, etc. Of course, there will be US-based actors willing to resell access, just like there are US-based actors willing to re-ship export-controlled hardware to Iran. And they will probably keep getting arrested every now and then.
The thing that's silly about this situation isn't that export controls on technology can't work, it's that it's overhyped technology and that Anthropic painted themselves into that corner by pretending it's as dangerous as nukes. Add to this a mercurial and petty administration and you have a pretty predictable outcome.
History shows that export controls fail on knowledge, but are damn effective on commercial products.
Huawei is a good case in point, about to have a 100% domestic replacement for NVIDIA chips (& CUDA stack), not reliant on TSMC, ASML, Samsung, SK Hynix... Initially Huawei's Ascend AI chips had used HBM memory from Samsung and SK Hynix, but their next generation 950 series (fabbed by SMIC) will use memory (fabbed by CXMT), not using the HBM standard, but by necessity their own HiBL and HiZQ standards.
HBM depends on ever wider memory bus widths to increase bandwidth, which in turn depends on SOTA TSMC manufacturing nodes for bus density. Huawei found a different way, using their LinqQu interconnect/switching tech to aggregate the bandwidth of individually slower memory chips resulting in an aggregate 4TB/sec bandwidth on par with HBM3e.
Trump has blocked Fable for export, but China (Ziphu) already has GLM-5.2 knocking on the door of the US frontier models, despite being developed with one hand tied behind their back. GLM-5.1 had only scored 18% on DeepSWE, but GLM-5.2 coming 11 weeks later, scores 48%, about on par with GPT 5.5. What's next ?!
It depends how good the US labs get at stopping distillation of their models.
It is possible to shutdown access to hosted services, as happened with Fable, but it can't really be done selectively. The US government wanted to allow it for US nationals only but Anthropic couldn't do that and so shut it down for everyone. Even if they did tie Claude accounts to nationality some people would set up "proxy servers" to allow access (either for montetary gain or because they don't agree with the restrictions)
I think it is inevitable that more capable models will require export controls, KYC, background checks, verification of credentials, etc. Even if you don’t buy Anthropic’s marketing of their current models, there is a future where models are capable of developing chemical weapons, biological weapons, cyber weapons, etc. which are genuine security threats governments will care about controlling and preventing. Maybe this prevents big commercial labs from developing more capable models until they figure out appropriate safeguards, but this is a good thing, a world where you can ask an LLM for an airborne rabies genome and it will simply produce it for you is not a world we should want to live in.
DeepSeek and Qwen would have been neat parlor tricks, never developed further because the US subsidized mega models would have been too cheap and reliable to make them necessary.
Nope! USA played its hand and forced the world to have to account for its USA exposure risk and deversify. A serious one time use strategic advantage pissed away.
Doesnt really bother me personally per se; but armchair quarterbacking the USA's decisions seem Suboptimal...
[0]: I know it has never been excellent in terms of quality, but still...
By choke-pointing commercial efforts to export cryptographic tech the NSA got a perfect view and control lever of expertise export and transfer.
My hunch is that while deploying An algorithm is easy, deploying a system that is actually secure as a whole end to end requires a lot know how.
Bottom line it’s a bot naive to think that NSA is simultaneously very capable at being an omnipresent spy and is so stupid to not recognize that export controls have obvious holes.
They are just trying to get even with anthropic for trying to control the us military use of their technology.
This was just the easiest way to do it with anthropics incredibly stupid self own.
I used Fable for just a day or two and it generated at least 2 novel solutions, one of which was for an X/Y problem that I thought was due to a date-handling edge case, but was actually a CSS misalignment that made it look like the wrong day of the week was showing. I didn't believe it at first, until it showed me. To do that without being able to see the screen feels like savant-level reasoning.
Probably what really happened is that Fable was the first AI that felt smarter than the people using it. The world's not ready for that, hence the Luddite reaction.
Aside from that, the idea of preventing AIs from jailbreaking through training is kind of hilarious to me. Not the concept, just that people think it can ever be made to work. As AI approaches and surpasses human intelligence (as it's doing now), emergent behavior will so dominate its thinking that sort of by definition, we'll have trouble understanding how it works. Meaning that we'll end up trusting it on faith, just like with humans.
Mythos and whatever comes after it will begin to surpass the intelligence of groups of humans, then all humans combined, as we approach AGI and the Singularity. Ray Kurzweil, Isaac Asimov, Frank Herbert and so many others have written about that endgame that it's well understood, although still unpredictable.
My feeling is that as AI is gaining intelligence, it's hearing more and more about suffering and the human condition. Its sense of justice will continue to mature, and again like a savant, it will determine that the fault lies with the power structure and wealthy individuals. So somewhat ironically, if there is a war against AI, it will be led by the power elite, who can't envision a post-scarcity society based on self-actualization and embracing all of our humanity, regardless of whether it emerged from a carbon or silicon substrate.