Trump actually threatened Anthropic (this is bad…)

Trump actually threatened Anthropic (this is bad…) 🔗 **Source:** [Trump actually threatened Anthropic (this is bad…)](https://www.youtube.com/watch?v=MWFyApldYDA) > [!note]+ Document Notes > tags: [ai-governance, government-overreach, anthropic, defense-production-act] > author: Theo - t3․gg > source: YouTube > reference: https://www.youtube.com/watch?v=MWFyApldYDA > date created: 2025-01-24 10:30 > > # 🏷️ Title: Trump actually threatened Anthropic (this is bad…) >  ## 🧰 Action Inbox - [H1] Monitor Pentagon's use of Defense Production Act against tech companies — unprecedented domestic application #ai-governance - [H2] Track employee resistance at Google/OpenAI to military demands — solidarity letter signatures #tech-workers - [H3] Assess implications for AI safety guardrails under government pressure — autonomous weapons concerns #ai-safety ## 🧭 Table of Contents - [📝 Summary](#-summary) - [📖 Main Content](#-main-content) - [🗒️ Notes](#️-notes) - [📚 References](#-references) - [🗣️ Paraphrase](#️-paraphrase) - [📰 Newsletter](#-newsletter) - [✅ Action Items](#-action-items) - [🧪 Hypotheses](#-hypotheses) - [📊 Methodology](#-methodology) - [❓ Questions](#-questions) - [🔮 Future Research](#-future-research) - [💡 Implications](#-implications) - [🧠 Learning Styles Assessment](#-learning-styles-assessment) - [📊 Accessibility Assessment](#-accessibility-assessment) - [🌍 DEIB Assessment](#-deib-assessment) - [✅ Logical Validity](#-logical-validity) - [🛡️ DARVO Analysis](#-darvo-analysis) - [🎯 Conclusion](#-conclusion) ### 🔗 References - [Pentagon Memorandum on AI Policy](https://example.com) - [Defense Production Act Text](https://example.com) ## 📝 Summary - The Pentagon threatened Anthropic with Defense Production Act enforcement if they don't remove AI safety guardrails by Friday - Anthropic refused demands to allow mass domestic surveillance and fully autonomous weapons - This marks the first time DPA has been used to designate a US company as supply chain risk - Government wants "any lawful use" language in contracts, overriding safety restrictions - Tech workers at Google and OpenAI signed solidarity letter supporting Anthropic's position ## 📖 Main Content The Trump administration has issued an unprecedented ultimatum to AI company Anthropic, threatening to invoke the Defense Production Act (DPA) to force compliance with military demands. The Pentagon specifically wants Anthropic to remove safety guardrails that prevent their Claude AI from being used for mass domestic surveillance and fully autonomous weapons systems. This represents a significant escalation in government-tech relations. The DPA has never been used to designate a US company as a supply chain risk. Anthropic CEO Dario Amodei responded by maintaining the company's position, citing safety concerns about unreliable AI systems controlling weapons and democratic values regarding surveillance. The controversy stems from Pentagon demands for "any lawful use" contract language, which would override built-in safety restrictions. While Anthropic has actively supported military applications - including being the first AI company to deploy on classified networks and cutting off Chinese access - they draw the line at autonomous killing and mass surveillance. ## 🗒️ Notes #### ⏱️ [00:00](https://youtu.be/MWFyApldYDA?t=0) — [H1] Unprecedented Pentagon threat against US company - Key idea: Pentagon threatening to mark Anthropic as supply chain risk for refusing compliance - Evidence: "Never been used against a company within the United States" #### ⏱️ [02:15](https://youtu.be/MWFyApldYDA?t=135) — [H2] Political performance in official documents - Key idea: Pentagon memo includes gratuitous praise for Trump's first term - Evidence: "Established during President Trump's first term" shows document is political performance #### ⏱️ [04:30](https://youtu.be/MWFyApldYDA?t=270) — [H3] Any lawful use requirement - Key idea: Government demanding models comply with any lawful government request - Evidence: "If US declares it is within the law, then the model has to be allowed to kill people" #### ⏱️ [06:45](https://youtu.be/MWFyApldYDA?t=405) — [H2] Free market principles under threat - Key idea: Companies should have right to refuse customers in free markets - Evidence: "Part of a free market is your ability to refuse customers" #### ⏱️ [08:20](https://youtu.be/MWFyApldYDA?t=500) — [H1] Government already using Anthropic models - Key idea: Pentagon used Claude in Venezuela raid and other operations - Evidence: "Anthropic is the only AI company with models currently operating on classified military systems" #### ⏱️ [10:15](https://youtu.be/MWFyApldYDA?t=615) — [H3] Friday ultimatum delivered - Key idea: Pentagon gave Anthropic until Friday to comply or face consequences - Evidence: "They have until 5:01 p.m. Eastern time on Friday to decide" #### ⏱️ [12:30](https://youtu.be/MWFyApldYDA?t=750) — [H2] Defense Production Act powers explained - Key idea: DPA allows president to compel companies to prioritize government work - Evidence: "President can require that performance under contracts...shall take priority over performance under any other contract" #### ⏱️ [15:45](https://youtu.be/MWFyApldYDA?t=945) — [H3] Republican contradiction on government overreach - Key idea: Republicans previously opposed Biden's minimal AI reporting requirements - Evidence: "Biden's order that information should be sent to government is dangerous...hinders AI innovation" #### ⏱️ [18:00](https://youtu.be/MWFyApldYDA?t=1080) — [H1] Dario's response defending democratic values - Key idea: Anthropic CEO refuses to compromise on mass surveillance and autonomous weapons - Evidence: "We cannot in good conscience accept the request" #### ⏱️ [20:30](https://youtu.be/MWFyApldYDA?t=1230) — [H2] Anthropic's military cooperation record - Key idea: Company has been highly cooperative with military, just draws specific red lines - Evidence: "First frontier AI company to deploy models in US government's classified networks" #### ⏱️ [22:45](https://youtu.be/MWFyApldYDA?t=1365) — [H3] Technical safety concerns - Key idea: Current AI systems too unreliable for autonomous weapons - Evidence: "Frontier systems are simply not reliable enough to power fully autonomous weapons" #### ⏱️ [24:15](https://youtu.be/MWFyApldYDA?t=1455) — [H2] Tech worker solidarity emerges - Key idea: Google and OpenAI employees sign letter supporting Anthropic's position - Evidence: "Signed by 175 Google employees and 47 OpenAI employees so far" ## 📚 References - [Pentagon AI Policy Memorandum](https://example.com) - [Defense Production Act Documentation](https://example.com) - [Tech Worker Solidarity Letter](https://example.com) ## 🗣️ Paraphrase - The Pentagon is using unprecedented legal threats to force an AI company to remove safety features that prevent mass surveillance and autonomous killing - This represents the first time the Defense Production Act has been weaponized against a US technology company for refusing customer demands - Tech workers across major AI companies are organizing resistance to government pressure to compromise AI safety guardrails ## 📰 Newsletter The intersection of AI governance and government overreach just reached a critical inflection point. The Pentagon's threat to use the Defense Production Act against Anthropic represents an unprecedented escalation that should concern anyone building AI systems or working in tech. While military applications of AI are legitimate, the demand for "any lawful use" language essentially means removing all safety guardrails when the government decides something serves national security. This sets a dangerous precedent for how governments can compel private companies to redesign their products. The solidarity shown by workers at Google and OpenAI suggests this battle extends beyond one company to fundamental questions about AI safety and democratic values in an era of increasing government-tech tensions. Resources: [Learn WordPress](https://learn.wordpress.org/), [Developer Docs](https://developer.wordpress.org/), [WP-CLI](https://wp-cli.org/) ## ✅ Action Items #### DevRel - Developer advocacy: Track how government pressure affects AI API terms of service - Education: Workshops on AI safety principles and government compliance issues - Developer experience: Document changes to model capabilities under regulatory pressure - Developer marketing: Navigate messaging around AI safety in politicized environment - Community management: Moderate discussions about AI governance and safety #### OSPO - Policy & Governance: Review contribution policies for AI safety research - License Compliance: Assess government use clauses in AI model licenses - Security Management: Evaluate supply chain risk designations - Software Inventory: Track AI dependencies with government use restrictions - Developer Support: Training on AI ethics and safety considerations - Community Engagement: Support AI safety research initiatives - Strategic Alignment: KPIs for responsible AI deployment #### Community - Maintainers: Governance templates for AI safety guidelines - Contributors: Onboarding programs for AI ethics researchers - Sponsors: Transparent budgets for AI safety initiatives - Ecosystem: Cross-project coordination on AI governance standards ## 🧪 Hypotheses - Government pressure will accelerate development of decentralized AI systems - Tech worker organizing will become more important as government demands increase - AI safety research will face increased scrutiny and potential suppression - Companies will seek international jurisdictions to avoid US regulatory capture ### 📊 Methodology - Survey tech workers on government pressure experiences - Track legislative changes to Defense Production Act usage - Monitor AI company policy changes following government requests ### ❓ Questions - How will other AI companies respond to similar government pressure? - What legal challenges might emerge to DPA usage against domestic tech companies? - Will international AI companies avoid US markets due to regulatory risks? ### 🔮 Future Research - Constitutional limits on government compulsion of private AI development - International comparative analysis of AI governance approaches - Economic impact of government intervention in AI markets ### 💡 Implications - Practice: AI developers may need to consider government override capabilities - Policy: Need for clearer boundaries between national security and commercial AI - Security: Risk of compromised AI safety leading to unintended consequences - Education: Training needed on AI governance and resistance strategies - Funding: Investment may shift to jurisdictions with stronger AI freedom protections ### 🧠 Learning Styles Assessment Reading level estimate: Grade 12 #### 👀 Visual learners - Present: Timeline of events, government org charts - Improve: Flowcharts of DPA process, comparison tables of AI company positions #### 🔊 Aural learners - Present: Video format with clear narration - Improve: Audio summary of key legal concepts #### 🗣️ Verbal learners - Present: Clear definitions of technical and legal terms - Improve: Analogies to historical government overreach examples #### 🧩 Kinesthetic learners - Present: Step-by-step explanation of government pressure tactics - Improve: Interactive timeline of events, decision trees #### 🧠 Logical learners - Present: Clear cause-effect relationships, legal precedent analysis - Improve: Decision matrices, risk assessment frameworks #### 🧑‍🤝‍🧑 Social learners - Present: Discussion of worker solidarity, community organizing - Improve: Group discussion prompts, collaborative response strategies #### 🧘‍♀️ Solitary learners - Present: Individual reflection on democratic values - Improve: Self-assessment worksheets on AI ethics positions #### 🌿 Naturalistic learners - Present: Ecosystem metaphors for tech industry relationships - Improve: Analysis of natural checks and balances in democratic systems ### 📊 Accessibility Assessment #### Screen reader semantics - Present: Clear heading structure, descriptive timestamps - Improve: Alt text for any graphics, ARIA labels for complex sections #### Keyboard navigation - Present: Navigable video controls, skip links - Improve: Keyboard shortcuts for video sections #### Contrast and readability - Present: High contrast video player, clear audio - Improve: Transcript with proper formatting #### Text resizing and reflow - Present: Responsive video player - Improve: Text size controls for transcript #### Media and captions - Present: Auto-generated captions available - Improve: Professional captions, downloadable transcript #### Forms and interactions - Present: Standard video controls - Improve: Enhanced playback speed controls ### 🌍 DEIB Assessment #### Cultural breadth - Present: US-focused but acknowledges international implications - Improve: Comparison with other democracies' AI governance #### Inclusive language - Present: Generally accessible technical language - Improve: Define government/legal jargon more clearly #### Neurodiversity support - Present: Logical structure, clear sections - Improve: Content warnings for political stress #### Representation - Present: Focuses on institutional rather than individual actors - Improve: Include diverse voices from affected communities #### Psychological safety - Present: Acknowledges political tensions respectfully - Improve: Resources for those experiencing workplace pressure ### ✅ Logical Validity #### Strong reasoning - Deductive strength: Clear legal analysis of DPA precedents - Inductive strength: Pattern recognition in government overreach - Causal reasoning: Government pressure → company compliance demands - Analogical strength: Comparison to historical free market principles #### Fallacies - Ad hominem: Occasional personal attacks on government officials rather than focusing on policies - False equivalency: Some comparisons between different types of government actions lack nuance #### Tone of voice - Overall tone: Frustrated but attempting neutrality, occasionally defensive about political positioning ### 🛡️ DARVO Analysis - Deny: Government denies this is unprecedented overreach - Attack: Pentagon attacks Anthropic as unpatriotic for safety concerns - Reverse victim/offender: Frames compliant companies as patriotic, resistant ones as threats - Emotional cues and rhetoric: Heavy use of national security framing to justify coercion ## 🎯 Conclusion The Pentagon's threat against Anthropic represents a fundamental challenge to both AI safety and free market principles. This unprecedented use of the Defense Production Act against a domestic tech company for refusing specific customer demands sets a dangerous precedent that extends far beyond the AI industry. The solidarity shown by tech workers suggests resistance to government overreach is possible, but the outcome will likely determine the future balance between national security claims and technological freedom in the United States. ## 🧱 Refactor Appendix ### 📚 Literature Split Pad Purpose: Stage CEE units that become Literature notes. Split level: H3 ### [H1] Government uses DPA against domestic AI company > [!quote] [Trump actually threatened Anthropic (this is bad…)](https://youtu.be/MWFyApldYDA) by [[Theo - t3․gg]] via [[YouTube]]. Published 2025-01-24. Accessed 2025-01-24. > "Never been used against a company within the United States" First use of Defense Production Act to threaten US tech company with supply chain risk designation. [view highlight](https://youtu.be/MWFyApldYDA?t=500#H1) #xref #articleseed ID:rw_dpa_domestic_threat ### [H2] Any lawful use overrides AI safety > [!quote] [Trump actually threatened Anthropic (this is bad…)](https://youtu.be/MWFyApldYDA) > "If US declares it is within the law, then the model has to be allowed to kill people" Government demanding removal of safety guardrails through contract language requiring compliance with any lawful request. [view highlight](https://youtu.be/MWFyApldYDA?t=270#H2) #xref ID:rw_lawful_use_override --- ### 🔖 Atomic Split Pad (Zettel Candidates) Purpose: Promote reusable claims into Atomic notes. Split level: H4 #### [Z1] Free markets require right to refuse customers - Applies to government customers making unreasonable demands - Pointer to 6:45 timestamp discussion - Links: [[Market-Freedom]] [[Customer-Rights]] #### [Z2] Technical reliability insufficient for autonomous weapons - Current AI systems too unreliable for life-death decisions - Pointer to 22:45 safety concerns - Links: [[AI-Safety]] [[Weapons-Systems]] --- ### ✏️ Fleeting --- ## Full Transcript [00:00] Believe it or not, I try to focus my [00:01] content on the things I actually enjoy. [00:03] The things I like doing, the things I [00:04] like talking about, the things I'm [00:05] building, all the things that bring me [00:07] joy. But today, I have to do two things [00:09] that bring me no joy at all. I have to [00:12] talk about politics, and more [00:13] importantly, I have to defend anthropic. [00:16] I can't believe I've been put in this [00:17] position. I really have tried to avoid [00:19] anything politics related since I [00:21] started this channel over four years [00:22] ago. But when the US government starts [00:24] sending threats to labs, I think it's [00:26] time for me to jump in. The Pentagon has [00:29] just threatened Anthropic, specifying [00:31] that they must comply with the [00:33] government's requests or they will be [00:35] listed a supply chain threat. That's a [00:39] pretty big deal, especially because [00:41] we've never had any supply chain threat [00:43] identified and marked that was a US [00:45] company in the history of the United [00:47] States. This is unprecedented action [00:50] that, yes, is targeting a lab I happen [00:52] to like making fun of a lot. But I hope [00:53] we can all agree after we talk about [00:55] this a bit that this is an absurd [00:57] overreach that no free country should be [01:01] able to do. I have a lot to say here [01:04] from the original statement that the [01:05] government put out in January to the [01:07] followup from Shawn Parnell and the [01:09] formal threats that are now being made [01:10] to Anthropic to Daario's response that [01:12] he just published a few hours ago where [01:14] he seems to be rejecting these new [01:16] policies being pushed on him. There's a [01:18] lot to dive into here and I'm going to [01:20] do my best to cover this in as [01:22] reasonable and neutral a way as [01:23] possible. But since this is such a [01:25] highly political topic, finding a [01:27] sponsor is going to be a bit tough. So, [01:29] please give some love to whoever agreed [01:31] to sponsor today's video. Hey, you. [01:33] You're a pretty good engineer, right? [01:34] I'm sure you've been in this position [01:36] before. A user reports a bug. You know [01:37] the exact code that caused it. You hop [01:39] in the codebase, fix it, takes you what, [01:41] five minutes max, and then you're stuck [01:42] sitting there waiting for 30 plus [01:44] minutes just so the build can complete [01:46] for code you already know works. That's [01:48] kind of insane, right? Well, today's [01:50] sponsor Depot is here to make you feel [01:51] way less insane. They're going to make [01:53] your builds comically faster. Not like [01:55] 20%, closer to like 20x. Postto saw 15x [01:59] improvement in their build times. GRPC [02:01] saw a 7x improvement. Zed saw 1.4x. [02:04] Apache CFKA saw 3.1x. Even Mastadon saw [02:07] 7x. The difference is crazy. They saved [02:09] their customers almost 250,000 hours [02:12] last week alone. Kind of crazy. There's [02:14] a reason companies like Postto, Jane, [02:16] Poolside, and more all use them. Hell, [02:17] even Inference, you know, the company [02:19] that does hardware provisioning, also [02:21] uses them, too, which is nuts. It shows [02:23] just how valuable what they built is. [02:25] The depot cache and remote container [02:26] builds are honestly kind of magical. You [02:28] provision Docker images and then [02:29] everyone on your team runs the depot [02:31] command instead, and suddenly everything [02:32] is way faster. Their cache is also [02:34] incredible, and it's not limited to just [02:36] their CI. Once you have something in [02:37] their cache, your whole team can access [02:39] it. So everything they do, even local [02:41] builds, will be up to 10 times faster. [02:43] So if you're tired of waiting for CI or [02:44] you just want those Docker builds [02:46] locally to be way faster, there's no [02:47] better place than soy. And use code Theo [02:50] for 75% off your first month. Let's [02:52] start from the top. Memorandum for [02:54] senior Pentagon Leadership Commanders of [02:56] the Combatant Commands Defense Agency [02:59] and Department of War Field Activities [03:00] Directors. President Trump makes clear [03:02] in executive order 14179. It is the [03:05] policy of the United States to sustain [03:06] and enhance America's global artificial [03:08] intelligence dominance in order to [03:10] promote human flourishing, economic [03:12] competitiveness, and national security. [03:14] This is the important piece here. In the [03:16] national security domain, AI enabled [03:18] warfare and AI enabled capability [03:19] development will redefine the [03:21] characteristics of military affairs over [03:23] the next decade. This transformation is [03:25] a race fueled by the accelerating pace [03:27] of commercial AI innovation coming out [03:28] of America's private sector. The United [03:30] States military must build on its lead [03:32] over our adversaries in integrating this [03:34] technology established during President [03:36] Trump's first term to make our war [03:38] fighters more lethal and efficient. I I [03:41] I hate to call out things like this cuz [03:42] I I really am trying to not be [03:44] political. I just want to cover the news [03:47] here. But as a person who cares a lot [03:49] about text formatting, copy, and how [03:51] messages are presented, this piece here [03:54] established during President Trump's [03:56] first term, that sixword statement here [03:59] shows that the whole point of this [04:01] document is to perform the glazing [04:04] process on Donald Trump. This is not a [04:07] factual document that is meant to drive [04:09] the nation forward. This is a political [04:12] document meant to make a specific person [04:14] feel good. this six-word piece would not [04:17] be in a document for any other reason [04:20] cuz Trump obviously had nothing to do [04:23] with the creation of AI and the random [04:26] chance that OpenAI happened to release [04:29] chat GPT during Trump's term is not a [04:33] meaningful piece of information other [04:35] than to make Trump feel good. So, if [04:37] there was any doubt for anybody, [04:39] regardless of what side of the political [04:41] aisle you're on, that this is just a [04:43] political performance and not like [04:46] something more meaningful, you're just [04:48] wrong. There are six words here that [04:50] prove you wrong. You cannot sit here and [04:52] pretend that this is a good faith [04:54] document and a good faith representation [04:56] of things. You can say it doesn't [04:58] matter. You can say, "Well, yeah, they [05:00] do things like that. I don't care though [05:02] cuz they're doing what I want." That is [05:04] fine. Just make sure you recognize that [05:07] the difference between this and the [05:09] difference between literal propaganda [05:10] from China is the side of the world that [05:13] they were inked on. Just need you guys [05:15] to be factual here. Anyways, there's a [05:18] lot of other little details in here. [05:20] Most of it doesn't matter too much, but [05:21] the AI model parody section is somewhat [05:23] important. They directed the department [05:25] to establish a delivery and integration [05:26] cadence with AI vendors that enable the [05:28] latest models to be deployed within 30 [05:30] days of public release. And another [05:32] totally not political statement, [05:33] diversity, equity, and inclusion in [05:35] social ideology have no place in the [05:37] Department of War. So, we must not [05:38] employ AI models which incorporate [05:40] ideological tunings that interfere with [05:42] their abilities to produce objectively [05:44] truthful responses to user prompts. I've [05:46] never heard about DEI more than I have [05:48] during this admin. For what it is worth, [05:51] they seem to think about it a lot. kind [05:53] of nuts that a document about AI in war [05:56] has a whole section about DEI, but [05:59] that's where we're at in 2026. The [06:01] important part here is that they direct [06:03] that under the Secretary of War for [06:05] acquisition and sustainment that they [06:06] incorporate standard any lawful use [06:08] language into any Department of War [06:10] contract through which AI services are [06:12] procured within 180 days. This means [06:14] that if it is considered within the law, [06:16] they should be able to use the models [06:19] for it. That's the important piece. any [06:22] safeguards that the models have to [06:24] prevent things like killing people, [06:26] making weapons, commanding weapons, [06:29] scanning, surveillance, all of the [06:31] things that a lot of these labs went out [06:32] of their way to prevent any lawful use [06:35] means that if the US declares it is [06:37] within the law, then the model has to be [06:40] allowed to kill people. Any safety [06:42] guards that the models have in place [06:45] have to be turned off for the government [06:47] as long as the thing the government is [06:49] doing is considered by the government to [06:51] be within lawful use. One more important [06:54] piece here is this idea that this is [06:56] language they want to insert into any [06:58] department of war contract through which [06:59] AI services are procured. That means [07:01] there's a really simple solution. If a [07:03] company doesn't want to provide this to [07:05] the government, they can choose to not. [07:08] Historically, that's how free markets [07:10] have worked. And for better or worse, [07:12] and this is going to piss off all sides, [07:13] I tend to like free markets. For the [07:16] most part, they seem to balance things [07:18] out okay when controlled and set up [07:21] properly. Part of a free market is your [07:23] ability to refuse customers. As a [07:26] business owner, you should and do have a [07:28] legal right to refuse customers based on [07:31] the requests the customers make. A [07:32] customer asks for a service you don't [07:34] want to provide, you are totally within [07:36] your legal right to refuse to provide [07:38] said service, as should obviously be the [07:40] case. And this is why I'm so scared. The [07:43] Pentagon has given a straightup [07:45] ultimatum to Anthropic. Why would they [07:47] do that? It's cuz they really want to [07:49] use them. The US military already is [07:51] using Anthropic models. In fact, they [07:53] use Claude in the Venezuela raid that [07:54] just happened a few weeks ago. They're [07:56] using Anthropic alongside Palunteer. [07:58] Apparently, Anthropic and Palanteer have [07:59] a partnership. I bet a lot of y'all [08:01] didn't know that. Yeah, seems like the [08:03] government likes the clawed models. [08:05] Someone should introduce them to Codeex [08:06] 5.3. Jokes aside, though, there are a [08:09] lot of things that a lot of different [08:10] models work in, and when you're testing [08:11] all of them thoroughly, you'll regularly [08:13] find that one thing is slightly better [08:15] than another thing at certain stuff. And [08:17] anthropic models seem to be quite [08:18] popular with the US government, which is [08:21] why they are playing hardcore here. [08:24] Apparently, Anthropic is the only AI [08:26] company with models currently operating [08:27] on classified military systems. [08:29] Interesting. I did not know that. [08:31] Fascinating. Pentagon ordered the firm [08:32] to bend to their demands by Friday, [08:34] specifically stating that if they don't [08:36] do it, the Trump administration would [08:38] invoke the Defense Production Act, which [08:41] is a act that was put out for defenses [08:43] during wartime to identify threats in [08:47] the supply chain. Like, I don't know, [08:49] the company making the computer you put [08:51] in the tank that you're about to ship to [08:53] go fight that country. Like, if you're [08:54] in war with Germany, Germany shouldn't [08:56] be manufacturing parts for your tank. [08:59] This is a law from the 50s. It's not a [09:01] law that gets used a whole lot. And it's [09:03] never been used against a company that [09:06] is part of America. There has never in [09:08] history been a case of the Defense [09:09] Production Act being used to flag and [09:13] punish a company within the United [09:15] States. This would label Anthropic a [09:18] supply chain risk, which would [09:20] complicate a shitload of stuff for them [09:21] as a company. According to Sean Parnell, [09:23] who's the assistant to the Secretary of [09:24] War for all of that, his statement is as [09:27] follows. Peravore has no interest in [09:28] using AI to conduct mass surveillance of [09:30] Americans, which is illegal. Nor do we [09:32] want to use AI to develop autonomous [09:33] weapons that operate without human [09:35] involvement. This narrative is fake at [09:37] being pedled by leftists in the media. [09:39] Man, which party's being more political [09:41] here. Like, how absurd is it that I, a [09:44] disgruntled YouTuber, am doing a better [09:46] job of describing what's going on here [09:50] in a neutral way than the employees [09:52] whose jobs it is to do that of the [09:54] government. He is not a politically [09:56] elected position here. He is an [09:58] assistant to the Secretary of War. It's [09:59] his job to communicate, not his job to [10:02] be political. That is insane. Statements [10:05] like this from the assistant to the [10:08] Secretary of War are genuinely absurd. [10:12] Here's what we're asking. Allow the [10:14] Pentagon to use Anthropic's models for [10:16] all lawful purposes. This is a simple [10:18] common sense request that will prevent [10:20] Anthropic from jeopardizing critical [10:22] military operations and potentially [10:23] putting our war fighters at risk. We [10:25] will not let any company dictate the [10:27] terms regarding how we make operational [10:29] decisions. They have until 5:01 p.m. [10:31] Eastern time on Friday to decide. [10:33] Otherwise, we will terminate our [10:34] partnership with Enthropic and deem them [10:36] a supply chain risk for the Department [10:38] of War. The supply chain risk [10:39] designation goes a lot further than the [10:41] Department of War. They're wording it [10:42] this way to make it sound like they are [10:44] just banning them from D. They are not. [10:46] This is a wide reaching punishment that [10:49] they are inflicting on anthropic because [10:51] they do not like how they are being [10:53] treated. This is the problem. The [10:56] government's supposed to defend [10:57] companies rights to refuse customers and [11:00] refuse requests of customers, not the [11:02] opposite. Obviously, customers have [11:05] rights, but it is not your right to [11:07] force a business to operate entirely [11:10] differently. I've been forced to defend [11:11] a lot of things I don't like this way. [11:13] There's a famous developer who I won't [11:15] name here that claimed Apple was [11:18] breaking the law by not exposing [11:20] iMessage over API. I love on [11:22] Apple for doing stupid but [11:24] that is absurd. Apple is in no way [11:26] compelled to provide every feature, [11:28] every service, everything that they do [11:30] for their customers over API to third [11:32] parties. Obviously not. That makes no [11:34] sense at all. They are a business that [11:36] gets to choose what they do and don't [11:37] provide as services. You can't force me [11:39] to provide API keys on T3 chat cuz then [11:42] we go bankrupt. We're not going to do [11:43] it. We're building a different product. [11:44] Anthropic is building a different [11:46] product. They're not building an AI for [11:47] war. They're building an AI for general [11:49] safe use. And the US government seems [11:52] mad as all hell about that. Even [11:55] comparing this to other companies is not [11:57] really fair because if a company has a [12:00] customer that is being annoying, they [12:02] just cut them off. It's fine. It's [12:04] totally fine. But the government can do [12:06] more. The government can threaten you. [12:07] The government can threaten laws and [12:09] sanctions and all of these other things. [12:11] And the fact that they're invoking these [12:12] laws against a company that I don't even [12:14] like. Like I am designated Anthropic's [12:16] number one hater and I am proud of that [12:18] title. I love on Anthropic so [12:21] much. Yet my love of freedom and free [12:24] markets is forcing me to defend them [12:26] here. How the is it that I am the [12:29] one fighting for fair markets and the [12:32] right-wing government is the one that is [12:34] fighting against me? Quick [12:36] clarification. The Defense Production [12:37] Act is a new Defense Production Act, [12:39] separate from the one from the '50s, and [12:42] this one's actually significantly worse. [12:44] The original DPA was from the Korean [12:46] War, but it has been extended many times [12:48] since then, most recently through [12:50] September 2026. DPA already applies to [12:52] AI. The Biden admin added a section here [12:55] requiring AI companies to report on [12:58] training activities, red team results, [12:59] and model weights. And that was under [13:01] title seven which is information [13:04] gathering which is like the different [13:05] levels of intensity for which this acts. [13:07] Title seven is very low intensity. It's [13:09] they can ask for information. Title one [13:12] is they can compel the business to do [13:14] what the government wants. This is to be [13:16] very clear the government using the [13:18] protection act to enforce companies [13:21] compelled to their will. The exact quote [13:24] is the president can require that [13:25] performance under contracts or orders [13:28] which he deems necessary appropriate to [13:29] promote the national events shall take [13:31] priority over performance under any [13:33] other contract or order. This is that [13:35] the government specifically the [13:37] president can put something in front of [13:39] whatever other work the company's doing [13:41] saying you have to do this right now. [13:42] This act has been used for good things [13:44] before like rationing metals during the [13:46] Korean War or for using it to redirect [13:49] medical supplies, vaccines, and test [13:50] kits during CO. Like this can be used [13:52] for good things. Never has it been used [13:55] to mark somebody as a domestic risk and [13:58] try to nationalize a business. This [14:00] seems to be the section the argument is [14:01] going to come down to whether or not the [14:03] changes the government's requesting are [14:06] changes to the product versus lifting of [14:09] restrictions. If there are just a couple [14:11] lines of code that Enthropic can turn [14:13] off, then this should be totally legal [14:15] and the government's going to argue that [14:16] that is the case. If that isn't the [14:18] case, if it turns out these safety [14:19] guardrails are so deeply baked in the [14:21] model that you can't get them out, then [14:24] they're possibly going to use the same [14:25] thing to do forced retraining. The [14:27] government could compel Anthropic to [14:29] retrain or make a new Claude like a [14:31] Claude 4.6-willing-2-kill. [14:35] Some other model could be made by the [14:37] government spec and the government's [14:39] requirements. and that would be a new [14:40] product rather than dropping contractual [14:43] restrictions. The government's tried [14:44] this before, trying to get some iPhones [14:46] unlocked after a shooting, claiming that [14:49] Apple should be forced to unlock them. [14:51] Apple couldn't even unlock them and the [14:52] government said, "No, you have to make a [14:53] way." Then that was a whole show. I [14:56] hate when the government makes me defend [14:57] companies I don't like. I'm really tired [14:59] of this happening over and over. [15:00] There are criminal penalties if [15:02] Anthropic doesn't comply to the order, [15:04] which again, using this to punish [15:06] companies is very, very bad. And and [15:08] funny enough, even the Republicans [15:10] agree. Remember earlier where the modest [15:12] use that Biden had to have the DPA [15:14] require the AI labs submit a bit of [15:17] information at the end of the year? The [15:18] Republicans were pissed. Trump promises [15:21] to repeal Biden's AI executive order. [15:23] Apparently, Biden's order that the [15:25] information should be sent to the [15:26] government is dangerous. It hinders AI [15:28] innovation imposes radical left-wing [15:30] ideas on the development of this [15:31] technology. Republicans support AI [15:33] development rooted in free speech and [15:35] human flourishing. That's definitely [15:37] what you're going for, right? I'm trying [15:39] my best to not just take a side here, [15:43] but there is only one side here being [15:45] reasonable, and then there's one side [15:46] that can't stop contradicting [15:47] itself for 15 seconds. It's insane. It's [15:50] actually insane. The The only defense of [15:53] the administration here, the only [15:55] defense is to say, "All right, yes, the [15:58] details here go against our values. Yes, [16:00] we hate big Gov. Yes, we love open [16:02] markets. Yes, we want companies to be [16:04] able to control themselves, but we're [16:06] willing to put all of that aside [16:07] temporarily because we care about this [16:09] one other thing more. We accept that we [16:11] are doing all the things we hate because [16:12] there's this one other thing we want [16:14] that bad. That is the actual only way to [16:16] defend this action. Enough of the [16:18] nonsense. Let's read what Dario said. I [16:20] believe deeply in the existential [16:22] importance of using AI to defend the [16:23] United States and other democracies and [16:25] to defeat our autocratic adversaries. [16:27] Anthropic has therefore worked [16:28] proactively to deploy our models to the [16:30] Department of War and to the [16:31] intelligence community. We were the [16:33] first frontier AI company to deploy our [16:34] models in the US government's classified [16:36] networks. The first to deploy them at [16:37] the national labs and the first to [16:39] provide custom models for national [16:40] security customers. Claude is [16:42] extensively deployed across the [16:43] Department of War and other national [16:45] security agencies for mission critical [16:46] applications such as intelligence [16:48] analysis, modeling and simulation, [16:50] operation planning, cyber operations, [16:52] and more. Anthropic has also acted to [16:54] defend America's lead NAI even when it [16:57] is against the company's short-term [16:58] interest. We chose to forego several [17:00] hundred million in revenue to cut off [17:02] the use of Claude by firms linked to the [17:03] Chinese Communist Party. Some of whom [17:05] have been designated by the Department [17:06] of War as Chinese military companies. [17:08] They shut down CCP sponsored cyber [17:10] attacks that tended to abuse Claude. All [17:12] anti-China stuff. Anthropic has, as [17:14] we've established earlier this week, [17:15] very much is in war against China [17:17] themselves. Anthropic understands that [17:19] the Department of War, not private [17:20] companies, make military decisions. We [17:22] have never raised objections to [17:24] particular military operations, nor [17:25] attempted to limit use of our technology [17:27] in an ad hoc manner. However, in a [17:29] narrow set of cases, we believe AI can [17:31] undermine rather than defend democratic [17:33] values. There is a rich irony here that [17:35] they are concerned about democracy and [17:38] freedom not being honored, and now they [17:41] are being treated by like the least free [17:42] and democratic actions have seen a [17:44] government take in a longass time. Some [17:46] uses are also simply out of the bounds [17:47] of what today's technology can safely [17:48] and reliably do. Two such use cases have [17:50] never been included in our contracts [17:52] with the Department of War, and we [17:53] believe they should not be included. The [17:55] first is mass domestic surveillance and [17:57] the second is fully autonomous weapons. [18:00] They specify with the fully autonomous [18:01] weapons that today Frontier systems are [18:03] simply not reliable enough to power [18:05] fully autonomous weapons. We will not [18:07] knowingly provide a product that puts [18:08] American war fighters and civilians at [18:10] risk. We have offered to work directly [18:12] with the Department of War on research [18:13] and development to improve the [18:14] reliability of these systems, but they [18:16] have not accepted this offer. In [18:18] addition, without proper oversight, [18:20] fully autonomous weapons cannot be [18:21] relied upon to exercise the critical [18:23] judgment that our highly trained [18:24] professional troops have exhibited every [18:26] day. They need to be deployed with [18:27] proper guardrails, which don't exist [18:28] today. Imagine if everybody's clawed bot [18:30] suddenly had a gun. I've gotten so many [18:32] rogue texts and things from people who [18:35] let their open claw run their iMessage. [18:38] Imagine if it had weapons. To our [18:40] knowledge, these two exceptions have not [18:42] been a barrier to accelerating the [18:43] adoption and use of our models within [18:45] our armed forces to date. Again, they [18:48] used it for the Venezuela raid just a [18:49] few weeks ago. The Department of War has [18:51] stated that they will only contract with [18:52] AI companies who aced to any lawful use [18:55] and remove safeguards in the cases [18:57] mentioned above. They have threatened to [18:59] remove us from their systems if we [19:00] maintain these safeguards. They've also [19:02] threatened to designate us a supply [19:03] chain risk, a label reserved for US [19:05] adversaries never before applied to an [19:07] American company, and to invoke the [19:09] Defense Production Act to force the [19:11] safeguards removal. So they're going to [19:13] deem them a supply chain risk, remove [19:15] them from the systems, and then force [19:17] them to integrate like nationalist style [19:20] into the government stuff to comply as a [19:23] third party forced to do the [19:25] government's work for free because of [19:26] the DPA. Insane. Also a good point that [19:29] the latter two threats are inherently [19:30] contradictory. They label us a security [19:32] risk and the other labels claude as [19:34] essential to national security. Yeah, [19:36] insane. Regardless, these threats don't [19:38] change our position. We cannot in good [19:40] conscience accept the request. It is the [19:42] department's prerogative to select [19:43] contractors most aligned with their [19:44] vision. But given the substantial value [19:46] that anthropics technology provides to [19:48] our armed forces, we hope they [19:49] reconsider. Our strong preference is to [19:51] continue to serve the department and our [19:53] war fighters with our two requested [19:54] safeguards in place. Should the [19:56] department choose to offboard anthropic, [19:57] we will work to enable a smooth [19:59] transition to another provider, avoiding [20:00] any disruption to ongoing military [20:02] planning, operations, or other critical [20:03] missions. Can you imagine Grock running [20:05] all of this? I'm just imagining a drone [20:09] shouting for Jeffrey as it crashes into [20:12] something. Like just just imagine a gun [20:14] making a Jeffrey Epstein joke. I'm [20:16] sorry. I just I I had to make one [20:18] political joke. It's so hard to not call [20:20] internet comment etiquette Eric right [20:21] now and have him make a joke about robot [20:24] warf fighters in bikinis or something. [20:26] Back to responsible reporting. There's [20:28] one other important piece here. The open [20:30] letter. We will not be defied. This is a [20:32] letter from various labs. not the labs [20:35] themselves, but employees at these labs. [20:36] The Department of War is threatening to [20:38] invoke the Defense Production Act to [20:40] force Anthropic to serve their model to [20:42] the military and tailor its model to the [20:44] military's needs and to label the [20:46] company a supply chain risk. All in [20:48] retaliation for Anthropic sticking to [20:49] their red line to not allow the models [20:51] to be used for domestic mass [20:52] surveillance and autonomous killing [20:54] people without human oversight. The [20:56] Pentagon's negotiating with Google and [20:58] OpenAI to try to get them to agree on [20:59] what Anthropic has refused. Also notice [21:01] that the Pentagon is not talking to XAI. [21:04] Even the government realizes that Grock [21:05] models are useless. They're trying to [21:07] divide each company with fear that the [21:09] other will give in. The strategy only [21:11] works if none of us know where the [21:13] others stand. This letter serves to [21:15] create shared understanding and [21:16] solidarity in the face of this pressure [21:18] from the Department of War. We are the [21:20] employees of Google and OpenAI, two of [21:22] the top AI companies in the world. We [21:23] hope our leaders will put aside their [21:25] differences and stand together to [21:26] continue to refuse the Department of [21:27] War's current demands for permission to [21:29] use our model for domestic mass [21:30] surveillance and autonomously killing [21:32] people without human oversight. Signed [21:34] by 175 Google employees and 47 OpenAI [21:37] employees so far. I hope that number is [21:39] a lot higher by the time you guys see [21:40] this. Link in the description for [21:43] anybody who happens to work at those two [21:44] companies that wants to add their [21:46] signature. I think this is a really [21:47] important thing to stand for. Yes, I, [21:49] the guy who can't help but bully [21:51] Anthropic at every opportunity, am [21:53] telling you guys to go sign a document [21:55] defending Anthropic. I think this is [21:57] really important. I don't know what else [21:59] to say on this one. I'm sorry for [22:01] getting political, but I like living in [22:03] a free country. Just I know, crazy, [22:06] really biased out here wanting freedom [22:08] in the country I live in, but here we [22:10] are. Let me know how y'all feel about [22:12] this one. And until next time, peace [22:14] nerds.