There doesn't really seem to be anything of substance in the actual executive order.
Section 1 doesn't say anything
Section 2 seems to boil down to: "improve cyber security and maybe use AI if we can find funding for it"
Section 3 proposes building a benchmark for evaluating cyber security performance of models that developers can choose to benchmark against. This seems like a good idea, I know Jack Clark has been a huge advocate for government's getting in with benchmarking.
Section 4 says to prioritize prosecuting cyber crimes. Not sure why they wouldn't already be prosecuted.
> Section 4 says to prioritize prosecuting cyber crimes. Not sure why they wouldn't already be prosecuted.
Not a whole lot of federal prosecutors. They're very selective about what gets pursued or not.
If they can't reliably build cases with a >90% success rate, it doesn't get prioritized. There's like <500 (federal) convictions per year on this whole area.
We hear about a few big famous ones in the news here, but most of it goes completely unenforced.
Not to mention quitting in droves because very many don't want to take these cases or otherwise to stand in court and explain why current admin is not bound by existing laws, court orders, the US constitution in general, or internationally recognized human rights etc.
There is no actual regulation in EO 14319. It only covers federal government purchasing and vendor management. No one is required to change the "ideology" of an LLM, although they might not be able to sell it to the government.
Does the First Amendment actually let the US government dictate the types of speech you're allowed to put in your LLM? I mean, a US government that's bound by the Constitution, obviously.
While that issue hasn't been specifically tested in court yet, the current interpretation of the First Amendment probably wouldn't allow the US government to dictate the types of speech you're allowed to put in your LLM. But federal government purchasing decisions aren't generally bound by the First Amendment. In other words, government officials can generally refuse to purchase your LLM services if they don't like the speech it outputs. So there's no real constitutional concern with this EO.
I'm not claiming that this EO is sensible or enforceable, just that it's not prima facie unconstitutional.
The specific text reads like a favor to Elon Musk's xAI, since "Truth-seeking" is the buzzword Elon Musk frequently used to talk about Grok:
Sec. 3
Unbiased AI Principles.
It is the policy of the United States to promote the innovation and use of trustworthy AI. To advance that policy, agency heads shall, consistent with applicable law and in consideration
of guidance issued pursuant to section 4 of this order, procure only those LLMs developed in accordance with the following two principles (Unbiased AI Principles):
(a) Truth-seeking. LLMs shall be truthful in responding to user prompts seeking factual information or analysis. LLMs shall prioritize historical accuracy, scientific inquiry, and objectivity, and shall acknowledge uncertainty where reliable information is incomplete or contradictory.
(b) Ideological Neutrality. LLMs shall be neutral, nonpartisan tools that do not manipulate responses in favor of ideological dogmas such as DEI. Developers shall not intentionally encode partisan or ideological judgments into an LLM's outputs unless those judgments are prompted by or otherwise readily accessible to the end user.
Might be fair to say it’s setting the tone, though, that if you use “woke” (subjectively defined) ideology in any of your company’s marketing, documentation, or other communications you won’t be considered for government contracts. That’s a major blow for any company given the naked corruption and grift coming from the current admin.
It is surprising to me American companies completely absent from the open model space, even though we have historically seen companies doing open source.
They aren't completely absent. Google keeps releasing Gemma models. Nvidia publishes Nemotron. Microsoft has their Phi series. IBM publishes Granite. Even OpenAI released a new open model (gpt-oss) less than a year ago.
I was going to link all of these, some are better than others, but they're all reasonably capable. A lot of these have versions that can run on modest hardware too. Granite was the most surprising I learned about recently, wasn't too good with Zed though.
I think that models like Granite are less known because they aren't clear leaders in any particular area. This obscurity is also another sign of how fast models are developing. If current Granite models had been released 4 years ago, they would have been astonishing breakthroughs at the time.
One of the main reasons why companies start new open source projects is because having a good open source option in a given category will usually push the market value of software in that category to $0, and this can be strategically valuable. For example, Google released Android as an open source operating system because they make their money from ads and data collection, not from selling operating system licenses. All the cell phone companies switched from Windows Mobile and Symbian to Android, which gave Google a ton of user data to sell.
For AI, the most profitable part of the value chain is selling inference. None of the big American companies want to release a leading edge model as open source because this would drive the price of inference to $0. Meanwhile, open source AI models are a huge strategic initiative for China. Having commodity Chinese models that are as good as the leading edge American models from 6 months ago forces the American companies to keep paying more and more money to train better and better models since the amount of time they can collect rent on a model they've previously trained is limited to 6 months.
No one open sources their core competencies, GitHub never open sourced their networked filesystem and Heroku never open sourced their dyno sandboxing code. They open source ancillary tools.
> Compounding the problem, labs in China often release dual-use capable models as open-weight. Once a model is open-weight, safeguards that do exist can be removed, making the model available to any state or non-state actor to use for malicious purposes, including the cyber and CBRN misuse those safeguards were built to prevent.
I loathe Anthropic. many companies don't contribute to open-source, but for one to be actively hostile to open-source, to the degree they're lobbying the government to ban it, is uniquely evil. at least these gatekeepers call themselves what they are.
scraping CoT won't stop the advance of Chinese models. neither will a US "ban" on using such models. at this point I'm cheering for DeepSeek or Qwen to catch up to Anthropic. I support anyone who releases open weights.
Is OpenAI significantly better so far regarding this, at least publicly? I'm increasing my LLM spend this weekend, and this could impact my decision. And I'll prioritize supporting open-weight models moving forward — already Chatgpt's censorship and surveillance dissuade from asking it genuinely helpful questions.
OpenAI seems marginally better. they did release gpt-oss-120b, which was decent at the time. but certainly not much better, and they seemed even more on board with fully disabling guardrails for Uncle Sam than Anthropic was. then again, rumor has it that Anthropic's AI selected that Iranian elementary school as part of Palantir's Project Maven pipeline, so..
I strongly recommend open-weight wherever you can. assume any data you pass to a closed model (including opinions or political positions you intimate) will be retained and analyzed in unfriendly ways, either now or ten years from now.
This entire year with the IPOs and now this is because there's a trillion dollars betting on AI and they all know they have no moat, there's no more training data and they're seeing diminishing returns on scaling anyway, and it's inevitable that smaller, open-source models will catch up and become competitive. It's a complete disaster, the tech industry is broken.
"The final text asks some AI companies to submit their powerful new models to a voluntary government review 30 days before releasing the products to the public, a pause that would give federal agencies some time to gauge what threats the products may pose to sensitive financial, national security and other computer systems."
How specifically does that review work? I want to give federal agency Opus 4.8 now, while 4.7 has been out for a while (leaving Mythos aside for now). They have 30 days to figure out whether it poses a threat.
How do you do that? Is there an eval for this and if there is why can't they just make it public? What is the agencies objective (but proprietary?) analysis here?
I seriously doubt even the government actually knows or has a real plan, let alone one actually related to security. If it's anything like their track record, they'll just be asking the AI about a topic related to their enemies (i.e. anyone opposed to them in any way) to see if it says anything remotely positive about them, or anything remotely critical of the regime or out of line with the regime's "alternative facts".
It's in the text of the order, it directs NIST to:
> develop and maintain a classified benchmarking process to assess the advanced cyber capabilities of AI models and determine the threshold at which an AI model should be designated a “covered frontier model” for the purposes of this order
> The final text asks some AI companies to submit their powerful new models to a voluntary government review 30 days before releasing the products to the public, a pause that would give federal agencies some time to gauge what threats the products may pose to sensitive financial, national security and other computer systems.
> An earlier draft of the order had called for a voluntary review as much as 90 days in advance, a provision that some AI industry officials had called too onerous, POLITICO reported last month.
A 90 days delay on the release of new models would have been insane. I guess I'm glad it's been revised at least on this specific point.
> A 90 days delay on the release of new models would have been insane. I guess I'm glad it's been revised at least on this specific point.
What would have made it "insane" exactly? The only argument I can imagine is that it gives non-US models (e.g. DeepSeek) a potential edge in the market during that time. But this potentially seems to be mitigated it being banned in the US anyway [0].
Given society seems to have developed just fine prior to the release of LLMs, I don't understand what the rush for more powerful and - potentially - more dangerous iterations of this technology is. If there is a legitimate reason that 90 days is somehow catastrophic, can someone ELI5?
Somewhere in all this it is crazy that the choice could be between a US company creating an AI that could doom civilization or letting China create the AI that dooms civilization. Do we want to be the first to "summon the demon" in our own fashion or let China manifest it first. Not saying this is the choice, but it would be a crazy dillema, albeit easy choice imo, if it was.
Because EO can get annoying to fight, companies would prefer to not fighting it. That's why these actions are to be remembered, companies will complain, but they will also comply.
No one should have to submit any published work to government review, even voluntary. This is a basic speech issue.
Absolutely no one would be okay with authors being 'encouraged' to submit their works to a 'voluntary' review by the feds to ascertain if their ideas are threatening. AI models are NO different.
Yeah, the order itself seems like a fairly reasonable response to Mythos level capabilities. It does solve one problem of the frontier labs, which is safely coordinating releases without hitting antitrust regulations. It also makes a bigger moat for incumbents.
So going forward expect US models to respond only in ways considered appropriate by the administration. If people thought models were producing slop before... lol.
You're absolutely right, abs-o-lutely, everybody says so. A lot, lot lot of people have been saying, you know they come to me and they say, "Mr. Claude, I can't believe the stuff I'm hearing, everybody is telling me he's right, is it true?" And I tell 'em, I say you're goddam right, that's what I say, but honestly folks, despite the negative press covfefe we've had a hell of a year, and that's really what it is with the nuclear folks, you can't trust em as far as you can throw em if you ask me, and believe me I've been throwing them around a LO<token limit exceeded>
Yea the details here really matter -
is this truly a politically neutral security review to determine impact and potentially prepare for it - that seems alright.
is this a review of "wokeness" in models and rejecting them if they don't align with the party views - this should not be allowed.
A politically neutral committee that decides what the review entails is what would happen in a true democracy and not a puppet oligarchy like we have today.
All neutrality has been aggressively neutered in every agency, or the target agency dismantled, in the last few months. An agency either supports the administrations political decisions wholly, or... well there's no "or" because an agency that doesn't won't remain an agency for very long.
No... executive orders are not laws, they can only command the federal government, not individuals or corporations. Meaning this is mostly pointless unless you're using models hosted by the government.
Executive orders aren’t laws (an important fact that should be repeated often and loudly). However, there’s probably room for the executive branch of the government to influence model hosts, as a major funder and consumer.
The judicidial branch, so the courts. The government would have to sue the corporation to try to get them to do something, at which point (hopefully) the judge would strike it down.
What courts? Look at all that's been happening over the past months. How much of it have the courts been able to meaningfully impact, vs what's still in effect?
This will be an important thing to check going forward, but I don't see why we would presume that they're going to be subverted in this way. Importantly, this is a completely different problem space from "slop" as such - there's plenty of Chinese models that implement their censorship almost entirely through guardrails on what topics they're willing to discuss.
> Most people would probably say that’s a good thing, if I read the tea leaves correctly.
I'm very pessimistic that this is about AI safety. I think it's probably more about giving the Trump administration leverage over AI companies. It will be able to coerce them into e.g. propagandizing or surveilling or similar or else they will risk the same kind of "regulatory oversight" that caused television networks to fire comedians who made jokes the regime didn't like.
If AI is going to be regulated those regulations should be debated in public and based upon the resulting laws passed by the legislative process and not determined by royal decree.
Whether that’s abuse or not I am not equipped to say with any confidence. I’d be curious to understand why you think this particular case is one of abusing executive authority and when an EO might not be such a case?
Well of course as you point out, EOs have gone from single digits, to double-digits, to thousands, and now down to hundreds per POTUS.
Contextually, I think it's a very reasonable (and commonly held, in the academic world) take that the EOs have also gotten far more legislative and legal. This is partly (but only partly) owing to administrative deference delegated by congress.
It's also somewhat specific to technological innovations, which some EOs have sought to occupy the field on before the lumbering process of congress can respond. And it's not limited to published EOs either, but many executive actions, especially in the White House OLC. This was very obvious during the W. Bush administration as regards the (Lotus Domino) email system in place at that time (which was the topic of my thesis, so it kinda serves as a temporal landmark in my consideration of this issue, but I do genuinely think it was a new frontier in executive overreach and obfuscation of interests in terms of how the White House has approached its interactions with the internet).
> I don't think the allowed shapes of formed metal should be regulated either.
I hear you. I collect knives as a hobby, and always have some kind of a cutting tool on me - they solve a surprising amount of little day-to-day problems (unpacking things bought in a shop being a prime example). I lost one of my folders to the UK border guard because, while a 6.5 cm blade was OK, they said its locking mechanism is illegal in the country. What's particularly funny is that I was actually trying to get back to France then - when entering, nobody asked about any knives. I never got that one back. :(
I wish I knew what the people who wrote this law thought. A folder without a locking mechanism is just as dangerous to others in violent scenarios, but way more dangerous for the user in typical EDC tasks. In Poland, there is no limit on the length of the blade nor on the locking mechanism. Technically, carrying an automatic foldable scythe or a zweihander is legal; you can't, however, carry a sword-cane or any other blade that is disguised as another item, like an umbrella. To put that all in perspective: in both countries, just like almost everywhere else in the developed world, the most lethal type of knife is the good old kitchen knife - ubiquitous, solid, with a tip ideal for thrusts, with a handle that protects the user's hand during the thrust, and so on. Such knives are generally not within the scope of knife-related laws.
So yeah, I don't get the logic behind the knife regulations at all. I'm not sure if completely dropping all of them is the way to go, but they would definitely benefit from a rational reevaluation. As an example, making the locking mechanism mandatory, instead of banned, would have no impact on knife-related deaths while allowing quite a few people each year to actually still have all their fingers.
I'm afraid a similar thing will happen with LLMs and later AIs. Regulators will "compromise" and focus on some kind of danger that's not entirely impossible, but also not very probable (assassins with blades in umbrellas...?), will fight for months over semantics, then pass the regulations to absolutely no visible effect - and the really dangerous uses will become either normalized or at least will move to the gray zone. The judiciary will try its best to apply existing laws to new situations, and in some cases, that will inevitably fail. We'll all deal with the consequences of these failures, unfortunately.
Section 1 doesn't say anything
Section 2 seems to boil down to: "improve cyber security and maybe use AI if we can find funding for it"
Section 3 proposes building a benchmark for evaluating cyber security performance of models that developers can choose to benchmark against. This seems like a good idea, I know Jack Clark has been a huge advocate for government's getting in with benchmarking.
Section 4 says to prioritize prosecuting cyber crimes. Not sure why they wouldn't already be prosecuted.
Section 5 doesn't say anything
Not a whole lot of federal prosecutors. They're very selective about what gets pursued or not.
If they can't reliably build cases with a >90% success rate, it doesn't get prioritized. There's like <500 (federal) convictions per year on this whole area.
We hear about a few big famous ones in the news here, but most of it goes completely unenforced.
And lately they seem to spend most of their time in courts trying to argue that immigrants don't deserve due process
This Executive Order is just an expansion of the existing censorship framework.
https://www.federalregister.gov/documents/2025/07/28/2025-14...
I'm not claiming that this EO is sensible or enforceable, just that it's not prima facie unconstitutional.
Sec. 3
Unbiased AI Principles.
It is the policy of the United States to promote the innovation and use of trustworthy AI. To advance that policy, agency heads shall, consistent with applicable law and in consideration
of guidance issued pursuant to section 4 of this order, procure only those LLMs developed in accordance with the following two principles (Unbiased AI Principles):
(a) Truth-seeking. LLMs shall be truthful in responding to user prompts seeking factual information or analysis. LLMs shall prioritize historical accuracy, scientific inquiry, and objectivity, and shall acknowledge uncertainty where reliable information is incomplete or contradictory.
(b) Ideological Neutrality. LLMs shall be neutral, nonpartisan tools that do not manipulate responses in favor of ideological dogmas such as DEI. Developers shall not intentionally encode partisan or ideological judgments into an LLM's outputs unless those judgments are prompted by or otherwise readily accessible to the end user.
Step 2: Complain about how the OSS/Chinese/whatever models are doing releases without approval
Step 3: Prohibit, because "safety" and "financial risks"(?)
So this is the door-shutting Altman et al have been pushing for eh?
https://deepmind.google/models/gemma/gemma-4/
https://developer.nvidia.com/ai-models#:~:text=NVIDIA%20Nemo...
https://www.microsoft.com/en-us/research/blog/phi-4-reasonin...
https://www.ibm.com/granite
https://openai.com/index/introducing-gpt-oss/
For AI, the most profitable part of the value chain is selling inference. None of the big American companies want to release a leading edge model as open source because this would drive the price of inference to $0. Meanwhile, open source AI models are a huge strategic initiative for China. Having commodity Chinese models that are as good as the leading edge American models from 6 months ago forces the American companies to keep paying more and more money to train better and better models since the amount of time they can collect rent on a model they've previously trained is limited to 6 months.
https://www.joelonsoftware.com/2002/06/12/strategy-letter-v/
Meta/Llama: "What am I, chopped liver?"
I thought the thing keeping inference above $0 was the hardware, and even if that were free there's still the tyranny of the Landauer Limit.
The notable exception is of course the google play services, which is also strategic (they control the OEMs with this, among other things).
And the drivers, but that's mostly not them I think (they could possibly have required open source drivers though)
"The world doesn't go round. It flips over!"
https://www.anthropic.com/research/2028-ai-leadership
scraping CoT won't stop the advance of Chinese models. neither will a US "ban" on using such models. at this point I'm cheering for DeepSeek or Qwen to catch up to Anthropic. I support anyone who releases open weights.
I strongly recommend open-weight wherever you can. assume any data you pass to a closed model (including opinions or political positions you intimate) will be retained and analyzed in unfriendly ways, either now or ten years from now.
He who controls the porn controls the universe. - Baron Amodei
How specifically does that review work? I want to give federal agency Opus 4.8 now, while 4.7 has been out for a while (leaving Mythos aside for now). They have 30 days to figure out whether it poses a threat.
How do you do that? Is there an eval for this and if there is why can't they just make it public? What is the agencies objective (but proprietary?) analysis here?
Train it dumb on "systems:, user:" prompt pairs.
Unleash on "system:, user:" prompt pairs.
Guess which you're providing for evaluation.
> develop and maintain a classified benchmarking process to assess the advanced cyber capabilities of AI models and determine the threshold at which an AI model should be designated a “covered frontier model” for the purposes of this order
For the same reason the CIA doesn't publish the Windows exploits it finds?
- https://web.archive.org/web/20260602130637/https://www.techn... - https://web.archive.org/web/20260520190620/https://fortune.c...
> An earlier draft of the order had called for a voluntary review as much as 90 days in advance, a provision that some AI industry officials had called too onerous, POLITICO reported last month.
A 90 days delay on the release of new models would have been insane. I guess I'm glad it's been revised at least on this specific point.
What would have made it "insane" exactly? The only argument I can imagine is that it gives non-US models (e.g. DeepSeek) a potential edge in the market during that time. But this potentially seems to be mitigated it being banned in the US anyway [0].
Given society seems to have developed just fine prior to the release of LLMs, I don't understand what the rush for more powerful and - potentially - more dangerous iterations of this technology is. If there is a legitimate reason that 90 days is somehow catastrophic, can someone ELI5?
[0] https://statetechmagazine.com/article/2025/04/these-states-h...
China, obviously.
Absolutely no one would be okay with authors being 'encouraged' to submit their works to a 'voluntary' review by the feds to ascertain if their ideas are threatening. AI models are NO different.
IMO this isn't much more egregious than the "stop woke AI" executive order he signed in July 2025 which explicitly regulated the "ideology" of LLMs
https://www.paulhastings.com/insights/client-alerts/presiden...
is this a review of "wokeness" in models and rejecting them if they don't align with the party views - this should not be allowed.
A politically neutral committee that decides what the review entails is what would happen in a true democracy and not a puppet oligarchy like we have today.
You left out the part containing the “barrels of money” incentive.
A lot more than you think, apparently
https://www.justsecurity.org/107087/tracker-litigation-legal...
https://www.bis.gov/press-release/biden-harris-administratio...
More regulated rather than unregulated (or very lightly regulated).
Most people would probably say that’s a good thing, if I read the tea leaves correctly.
I'm very pessimistic that this is about AI safety. I think it's probably more about giving the Trump administration leverage over AI companies. It will be able to coerce them into e.g. propagandizing or surveilling or similar or else they will risk the same kind of "regulatory oversight" that caused television networks to fire comedians who made jokes the regime didn't like.
(probably a good thing, in this particular case)
must be TACO Tuesday
So if we’re going to be rational about it, I think it is better to critique the substance of the EO rather than its mere existence, which is common practice: https://www.presidency.ucsb.edu/statistics/data/executive-or...
So in that spirit, what do you think of the substance?
Yeah, and I hated that move in the exact same way I hate the one this thread is about.
How does that make it better for the current administration to do it?
Whether that’s abuse or not I am not equipped to say with any confidence. I’d be curious to understand why you think this particular case is one of abusing executive authority and when an EO might not be such a case?
Contextually, I think it's a very reasonable (and commonly held, in the academic world) take that the EOs have also gotten far more legislative and legal. This is partly (but only partly) owing to administrative deference delegated by congress.
It's also somewhat specific to technological innovations, which some EOs have sought to occupy the field on before the lumbering process of congress can respond. And it's not limited to published EOs either, but many executive actions, especially in the White House OLC. This was very obvious during the W. Bush administration as regards the (Lotus Domino) email system in place at that time (which was the topic of my thesis, so it kinda serves as a temporal landmark in my consideration of this issue, but I do genuinely think it was a new frontier in executive overreach and obfuscation of interests in terms of how the White House has approached its interactions with the internet).
you’re being so reductive you’ve made any discussion about it completely useless
If you do something bad with your tool (knife or LLM), though, that's the problem. And we have laws for that already.
I hear you. I collect knives as a hobby, and always have some kind of a cutting tool on me - they solve a surprising amount of little day-to-day problems (unpacking things bought in a shop being a prime example). I lost one of my folders to the UK border guard because, while a 6.5 cm blade was OK, they said its locking mechanism is illegal in the country. What's particularly funny is that I was actually trying to get back to France then - when entering, nobody asked about any knives. I never got that one back. :(
I wish I knew what the people who wrote this law thought. A folder without a locking mechanism is just as dangerous to others in violent scenarios, but way more dangerous for the user in typical EDC tasks. In Poland, there is no limit on the length of the blade nor on the locking mechanism. Technically, carrying an automatic foldable scythe or a zweihander is legal; you can't, however, carry a sword-cane or any other blade that is disguised as another item, like an umbrella. To put that all in perspective: in both countries, just like almost everywhere else in the developed world, the most lethal type of knife is the good old kitchen knife - ubiquitous, solid, with a tip ideal for thrusts, with a handle that protects the user's hand during the thrust, and so on. Such knives are generally not within the scope of knife-related laws.
So yeah, I don't get the logic behind the knife regulations at all. I'm not sure if completely dropping all of them is the way to go, but they would definitely benefit from a rational reevaluation. As an example, making the locking mechanism mandatory, instead of banned, would have no impact on knife-related deaths while allowing quite a few people each year to actually still have all their fingers.
I'm afraid a similar thing will happen with LLMs and later AIs. Regulators will "compromise" and focus on some kind of danger that's not entirely impossible, but also not very probable (assassins with blades in umbrellas...?), will fight for months over semantics, then pass the regulations to absolutely no visible effect - and the really dangerous uses will become either normalized or at least will move to the gray zone. The judiciary will try its best to apply existing laws to new situations, and in some cases, that will inevitably fail. We'll all deal with the consequences of these failures, unfortunately.
Plenty of innovations are regulated (ie, its regularity maintained) without the state.
Do we really imagine that intervention by the imperial hegemon is likely to lead to regulation, rather than capture and weaponization?