Decoder with Nilay Patel

Anthropic doesn't trust the Pentagon, and neither should you

March 12, 2026

Key Takeaways Copied to clipboard!

  • The core conflict between Anthropic and the Pentagon centers on Anthropic's refusal to participate in mass surveillance, stemming from a deep historical distrust of the US government's reinterpretation of surveillance laws like Executive Order 12333 and the Patriot Act. 
  • The historical expansion of US government surveillance capabilities, driven by clever legal interpretations by administrations of both parties since the Reagan era, has created a pattern where the plain English meaning of laws is overridden by secret or self-serving interpretations, as exemplified by the NSA's definition of 'target'. 
  • The current dispute highlights a clash between the blunt, unsubtle demands of the Trump administration for unrestricted access and Anthropic's public persona as a thoughtful, safety-conscious AI provider, further complicated by the argument that forcing Anthropic to build surveillance tools constitutes compelled speech under the First Amendment. 

Segments

Anthropic vs. Pentagon Legal Battle
Copied to clipboard!
(00:01:17)
  • Key Takeaway: Anthropic sued the Department of Defense after being designated a supply chain risk, alleging violations of its First and Fifth Amendment rights.
  • Summary: The Pentagon designated Anthropic a supply chain risk, leading Anthropic to file a lawsuit challenging the designation. Anthropic claims this action violates its First and Fifth Amendment rights and seeks to destroy the company’s economic value. The host notes this legal battle will be covered extensively in the future.
History of US Surveillance State
Copied to clipboard!
(00:02:01)
  • Key Takeaway: Government lawyers historically twist legal interpretations, such as the word ’target,’ to expand surveillance powers beyond what a plain reading of the law suggests.
  • Summary: The discussion pivots to the history of US surveillance, focusing on how government lawyers reinterpret simple words to expand surveillance authority, often only revealed through major controversies like the Snowden leaks. This pattern of expanding power is seen as continuing, with AI poised for the biggest expansion yet.
Guest Introduction and Expertise
Copied to clipboard!
(00:02:15)
  • Key Takeaway: Mike Masnick, CEO of Techdirt, is an expert on government overreach and the interconnected growth of the internet and the surveillance state over two decades.
  • Summary: Mike Masnick joins the ‘Decoder with Nilay Patel’ episode to provide context on digital privacy and surveillance. He has decades of experience writing about government overreach and the evolution of the surveillance state.
Surveillance Law vs. Government Action
Copied to clipboard!
(00:02:32)
  • Key Takeaway: There is a significant gap between what surveillance laws explicitly state, what the government actually intends to do, and how the government interprets those laws.
  • Summary: The government’s actions regarding surveillance often contradict the plain English meaning of the laws governing them. Government lawyers frequently engage in motivated reasoning to justify expanded surveillance, often succeeding because the process occurs in secret without adversarial challenge.
Patriot Act and FISA Court Context
Copied to clipboard!
(00:06:26)
  • Key Takeaway: Post-9/11 surveillance expansion was codified by the Patriot Act and facilitated by the secret, one-sided FISA court, which historically rubber-stamped over 99% of applications.
  • Summary: The Patriot Act granted surveillance abilities under the guise of preventing terrorism, while the FISA court, operating in secret with only the government pleading its case, allowed for broad interpretations. This system enabled the intelligence community to operate with limited oversight.
Executive Order 12333 and Data Collection
Copied to clipboard!
(00:07:27)
  • Key Takeaway: Executive Order 12333 allows the NSA to tap foreign communications, which inadvertently includes collecting US person communications if the data briefly leaves US territory en route.
  • Summary: EO 12333, established under Ronald Reagan, governs intelligence collection rules. The interpretation allows the NSA to collect data passing outside the US, even if the communication is entirely between two US persons. This collected data can be searched later, leading to a form of mass surveillance on Americans despite public claims to the contrary.
The ‘Target’ Redefinition
Copied to clipboard!
(00:09:18)
  • Key Takeaway: The NSA redefined ’target’ to mean that any communication mentioning a foreign person is fair game for collection, even if the communication itself is between two US persons.
  • Summary: The NSA’s interpretation allows them to collect communications of US persons if those communications mention or hint at a non-US person. This broad definition effectively turns communications between two Americans into potential surveillance targets if a foreign entity is mentioned.
Bipartisan Growth of Surveillance State
Copied to clipboard!
(00:12:14)
  • Key Takeaway: The expansion of the surveillance state has been an incremental process involving administrations and Congresses from both Democratic and Republican parties.
  • Summary: The growth of surveillance powers is not attributable to a single party, spanning from the Reagan era to the post-9/11 Bush administration and subsequent Obama administration actions. Presidents often acquiesce to intelligence community requests driven by the fear of another terrorist attack, allowing lawyers to bend legal interpretations.
Anthropic’s Red Lines vs. OpenAI’s Stance
Copied to clipboard!
(00:04:46)
  • Key Takeaway: Anthropic established specific red lines against autonomous weapons and mass surveillance, whereas OpenAI initially agreed to ‘all lawful uses,’ suggesting a failure to grasp the government’s redefined legal interpretations.
  • Summary: Anthropic explicitly drew lines against mass surveillance, distrusting the government’s meaning of ’lawful uses.’ In contrast, Sam Altman of OpenAI initially accepted ‘all lawful uses,’ which the guest suggests implies either ignorance of the reinterpreted laws or an attempt to play the same legal games as the NSA.
Trump Era Bluntness and Reckoning
Copied to clipboard!
(00:24:35)
  • Key Takeaway: The current administration’s lack of subtlety in announcing intentions, rather than relying on complex legal maneuvering, might finally force a public reckoning on surveillance.
  • Summary: Unlike previous administrations that used subtle lawyering to justify secret surveillance, the current administration is openly stating its desires, such as declaring a war footing with Iran. This bluntness, combined with Anthropic’s principled stand, could be the catalyst for a necessary public debate on surveillance.
Third-Party Doctrine and Cloud Data
Copied to clipboard!
(00:30:51)
  • Key Takeaway: The decades-old third-party doctrine allows the government to obtain data held by corporations (like cloud providers) without a warrant, effectively swallowing much of the Fourth Amendment’s protection.
  • Summary: The third-party doctrine dictates that data voluntarily shared with a third party, like phone records, is not protected by the Fourth Amendment against unreasonable search. Since nearly all digital activity is recorded by third parties (Amazon, Apple, etc.), the government can easily request this data without a warrant, bypassing constitutional protections.
Anthropic’s Role in Data Protection
Copied to clipboard!
(00:36:21)
  • Key Takeaway: Anthropic’s refusal to process data acquired by the government from third parties mirrors past corporate stands, like Apple’s refusal to build an FBI backdoor, by inserting a process barrier against mass surveillance.
  • Summary: Anthropic specifically objected to using Claude on third-party data collected by the government, which is the mechanism for mass surveillance. This stance aligns with large tech companies historically refusing government demands that cross ethical or legal lines, though the government’s reaction—a supply chain risk designation—is an extreme escalation.
Compelled Speech Argument
Copied to clipboard!
(00:39:28)
  • Key Takeaway: The argument that forcing Anthropic to build tools it objects to constitutes compelled speech under the First Amendment is considered a valid, though perhaps secondary, legal challenge.
  • Summary: FIRE advocates argue that compelling Anthropic to write code for surveillance tools it opposes is a violation of free speech rights, as code is considered speech. This mirrors past First Amendment claims raised when the government sought to force companies to build backdoors into encrypted systems.
The Verge’s AI Coverage Strategy
Copied to clipboard!
(00:42:03)
  • Key Takeaway: The Verge covers AI as a standalone beat, but uniquely roots its policy and business coverage in rigorous product testing to determine if the technology actually works for consumers.
  • Summary: The Verge is investing heavily in AI coverage across its policy and product desks, hiring specialized reporters. The publisher emphasizes that their unique strength lies in grounding the business and policy narratives in the actual consumer experience of using the products.
Supporting The Verge’s Journalism
Copied to clipboard!
(00:43:41)
  • Key Takeaway: Individual support through subscriptions is the single most important action readers can take to sustain The Verge’s independent coverage, which is protected by a strict firewall between editorial and business operations.
  • Summary: The Verge maintains a strong ethics policy ensuring coverage is not for sale, relying on advertising and subscriptions for support. Subscribing provides ad-free podcasts and access to specialized newsletters like ‘Regulator’ and ‘Optimizer,’ directly funding the expansion of their reporting staff.