The Architecture of Anticipation: Historical Lessons from Post-9/11 Surveillance for Artificial Intelligence Governance

Report body

The Architecture of Anticipation: Historical Lessons from Post-9/11 Surveillance for Artificial Intelligence Governance

The immediate aftermath of the September 11, 2001 attacks catalyzed a profound recalibration of the balance between state security and individual privacy in the United States and across the globe. In an environment defined by acute national trauma, the fear of subsequent catastrophic events, and the asymmetric nature of modern terrorism, legislatures rapidly enacted sweeping surveillance authorities. The Uniting and Strengthening America by Providing Appropriate Tools Required to Intercept and Obstruct Terrorism Act of 2001, commonly known as the USA PATRIOT Act, instantiated legal and technological architectures that fundamentally altered the relationship between the citizen and the state1. Two decades later, a strikingly similar narrative of unprecedented, potentially existential vulnerability is emerging around the rapid advancement of Artificial Intelligence (AI) and Artificial General Intelligence (AGI).
As governments and multinational corporations confront the dual-use capabilities of advanced generative models—ranging from the automated proliferation of bioweapons to the destabilization of digital information ecosystems—proposals to restrict, monitor, classify, filter, and regulate information access have gained immense political and institutional traction3. The mechanisms currently proposed for AI governance, including mandatory identity verification for compute access, centralized query logging, algorithmic content classification, and automated behavioral risk scoring, mirror the infrastructural logic of post-9/11 counterterrorism programs. In both eras, the underlying justification posits that the stakes are too high, and the threats too rapid, to rely on traditional, individualized, retroactive legal frameworks.
This exhaustive report provides a detailed historical comparison between the expansion of surveillance and government authority following the 2001 attacks and contemporary proposals to regulate artificial intelligence in the name of safety. By analyzing the trajectories of the USA PATRIOT Act, related surveillance authorities, subsequent disclosures, court controversies, and oversight reforms, this analysis identifies the structural patterns of emergency legislation. It subsequently investigates the strongest arguments for emerging AI information-control mechanisms, establishing exactly which similarities and differences are supported by evidence. Finally, this report defines the specific institutional safeguards—including warrants, minimization, adversarial proceedings, decentralization, and technological separation of identity from inquiry—necessary to prevent AI-era safety measures from calcifying into permanent infrastructures for monitoring lawful thought and inquiry.

Part I: The Anatomy of Post-9/11 Surveillance Expansion

The expansion of surveillance capabilities in the early 21st century was characterized by the rapid passage of complex legislation, the deployment of broad statutory terminology, the reliance on secret judicial interpretations, and the systematic circumvention of traditional Fourth Amendment constraints. To understand the risks inherent in contemporary AI regulation, one must first dissect the legal and institutional mechanisms that enabled the post-9/11 intelligence apparatus, paying specific attention to the divergence between public perception and institutional execution.

Emergency Conditions and Broad Terminology

The USA PATRIOT Act was enacted on October 26, 2001, with extraordinary speed, minimal congressional debate, and virtually no comprehensive prior study of its long-term constitutional implications1. The legislation expanded all traditional tools of surveillance used by law enforcement, including wiretaps, search warrants, pen register/trap-and-trace orders, and subpoenas, while simultaneously blurring the longstanding legal distinction between domestic criminal investigations and foreign intelligence gathering2. Prior to the Act, foreign intelligence gathering required that intelligence be "the purpose" of an investigation; the Patriot Act lowered this threshold to merely "a significant purpose," fundamentally entangling domestic law enforcement with foreign surveillance2.
A defining characteristic of this legislative expansion was the use of intentionally broad terminology. Section 215 of the USA PATRIOT Act, which amended the Foreign Intelligence Surveillance Act (FISA), authorized the Federal Bureau of Investigation (FBI) to compel the production of "any tangible things" (including books, records, papers, documents, and other items)7. Prior to this amendment, business record demands were restricted to specific entities such as common carriers, storage facilities, and vehicle rental agencies8. The new statutory language simply required the government to certify that the requested items were "relevant" to an authorized investigation to protect against international terrorism or clandestine intelligence activities9.

The Illusion of Targeted Surveillance: Public Belief vs. Classified Interpretation

The public discourse surrounding Section 215 largely focused on its potential application to specific, individualized records. The provision became infamously known as the "library provision" because civil liberties groups and the American Library Association fiercely warned that the FBI could use it to demand the borrowing records of individual citizens, thereby chilling First Amendment rights11. Citizens and lawmakers alike believed they were authorizing a tool that, while broad, would be used to investigate specific suspects by looking at their specific transactional histories.
In reality, the term "relevant," a standard legal concept traditionally used to define the scope of a subpoena based on a specific nexus to a known suspect, underwent a radical, classified reinterpretation. Operating entirely in secret, the Foreign Intelligence Surveillance Court (FISC) accepted the executive branch's argument that "relevance" could be applied prospectively to the bulk collection of the entire nation's telephony metadata9. The government's logic, endorsed by the FISC, posited that in order to find the records of suspected terrorists (the "needle"), it was mathematically necessary to possess the entire database of all domestic call detail records (the "haystack"). Therefore, every phone record in the United States was deemed "relevant" to counterterrorism investigations7.
This interpretation remained hidden from the American public, and indeed from many members of Congress, for over a decade15. The bulk telephone records program was repeatedly reauthorized by the FISC, effectively creating a massive body of secret law that authorized a domestic surveillance program bearing almost no resemblance to the original legislative intent of Section 2159. As the Privacy and Civil Liberties Oversight Board (PCLOB) later concluded, redefining the word "relevant" to encompass all calling records across the nation was circular, unlimited in scope, and out of step with traditional legal standards9.

National Security Letters, Gag Orders, and the Normalization of Exceptional Powers

While Section 215 required the nominal approval of the FISC, other expanded authorities bypassed the judiciary entirely. The USA PATRIOT Act vastly expanded the FBI's ability to issue National Security Letters (NSLs)—administrative subpoenas compelling telecommunications providers, internet service providers, and financial institutions to turn over customer metadata without prior judicial approval17.
The defining feature of the NSL is its accompanying non-disclosure provision, or "gag order," which prevents the recipient from disclosing the existence of the demand to anyone, including the target of the investigation or the media17. These gag orders operated as indefinite prior restraints on speech, effectively shielding the government's surveillance activities from public scrutiny and political accountability18. The constitutionality of these gag orders was repeatedly challenged, with courts initially striking them down as abhorrent to democracy, though appellate courts later upheld modified versions under strict national security premises17.
The lack of initial judicial oversight predictably resulted in severe mission creep and widespread abuse. Reports by the Department of Justice's Office of the Inspector General (OIG) revealed that the FBI issued over 143,000 NSL requests between 2003 and 2005 alone20. Furthermore, the OIG documented the pervasive use of "exigent letters"—informal requests for data based on factually inaccurate claims of emergency conditions, used to bypass even the minimal internal requirements of the NSL process21. This institutional behavior demonstrated a clear pattern: powers granted for the most exceptional circumstances (e.g., ticking time bombs) are rapidly normalized and integrated into routine investigative workflows, far removed from their original national security justifications.

Mission Creep and Parallel Construction: The Laundering of Intelligence

The expansion of surveillance infrastructure was not limited to the accumulation of data; it deeply infected the prosecutorial system through a practice known as "parallel construction," demonstrating a massive expansion from exceptional targets to wider populations. The Drug Enforcement Administration (DEA) established the Special Operations Division (SOD), which functioned as a clearinghouse to distribute highly classified intelligence—often collected via warrantless NSA surveillance programs under FISA—to domestic law enforcement agencies22.
Because the original sources of this intelligence were classified and legally dubious in a domestic context, the SOD instructed local law enforcement to conceal the true origins of the information. Officers were directed to construct a separate, parallel evidentiary basis for an arrest—such as orchestrating a pretextual traffic stop to discover narcotics that the NSA surveillance had already alerted them to25.
Parallel construction represents the ultimate subversion of the adversarial legal process. By obscuring the source of the evidence, the government entirely circumvents the exclusionary rule, denying criminal defendants their constitutional right to challenge the legality of the original surveillance under the Fourth Amendment26. This mechanism ensured that the most expansive surveillance programs remained insulated from appellate review and public constitutional scrutiny, demonstrating how infrastructure built for counterterrorism inevitably bleeds into routine domestic policing.

Chilling Effects and the Targeting of Inquiry

The realization that everyday interactions were subject to state monitoring generated profound "chilling effects" on lawful behavior. When individuals know, or even suspect, that their intellectual inquiries, reading habits, and research are being algorithmically sifted by intelligence agencies, they self-censor. A post-Snowden study by PEN America confirmed this phenomenon, documenting that over 500 writers surveyed were modifying their online behavior, avoiding certain search terms, and curtailing their exploration of controversial topics due to the perception of omnipresent surveillance28.
This dynamic validates the early concerns of the library associations during the Section 215 debates. Librarians argued that the mere potential for the FBI to review reading records would turn inquiring minds away from exploring varied avenues of thought, fundamentally eroding First Amendment protections29. The surveillance of intellectual inquiry directly undermines democratic participation, freedom of expression, and civic trust, proving that actual enforcement is not required to damage democratic freedoms; the perception of monitoring is sufficient30.

Institutional Incentives and the Illusion of Reversal

The architecture of post-9/11 surveillance relied heavily on the "Third-Party Doctrine," a legal principle holding that individuals have no reasonable expectation of privacy in information they voluntarily turn over to third parties, such as banks or telephone companies7. The surveillance state did not physically tap every phone line; rather, it mandated that private telecommunications companies act as extensions of the state's intelligence apparatus, establishing a critical precedent for the outsourcing of mass surveillance7. The government capitalized on institutional incentives, eventually granting telecommunications providers retroactive immunity for their participation in warrantless surveillance programs, effectively fusing corporate data collection with state security imperatives32.
Once surveillance infrastructure is entrenched, reversing it is exceptionally difficult. Following the 2013 disclosures by Edward Snowden, public outrage forced Congress to act9. The USA FREEDOM Act of 2015 was championed as a monumental reform that banned the government's bulk collection of domestic phone records13. However, the reality of the reform demonstrated the resilience of surveillance architectures. While the government could no longer hold the bulk database on its own servers, the USA FREEDOM Act mandated that the telecommunications companies retain the data, allowing the NSA to query the corporate databases using specific selection terms and to receive contact-chaining results ("hops") encompassing vast networks of individuals13. The core capability—the monitoring of associative metadata on a massive scale—was not dismantled; it was merely decentralized into the private sector.

Part II: The Emergence of AI and Information-Control Mechanisms

Two decades after the passage of the USA PATRIOT Act, a highly analogous paradigm is taking shape. The rapid evolution of Artificial General Intelligence (AGI) and powerful Large Language Models (LLMs) has prompted deep anxieties regarding systemic disruption, biosecurity, automated cyber warfare, and existential risk3. To mitigate these hazards, policymakers and technology companies are constructing a new infrastructure of control that heavily borrows from the post-9/11 surveillance playbook.

The Shift from Spatial to Cognitive Surveillance

Whereas earlier surveillance programs monitored who an individual communicated with (telephony metadata) and where they went (cell site location information), emerging AI models process, generate, and record the raw material of human cognition31. As legal scholars note, an AI chat log functions as a "personal journal" that users treat as an intimate sounding board for their fears, legal questions, medical symptoms, and intellectual explorations34.
The regulatory mechanisms currently proposed to constrain AI risks present profound threats to civil liberties precisely because they seek to monitor and filter this cognitive layer. An examination of these emerging mechanisms reveals striking parallels to the past.

Mandatory Identity and Age Verification

To prevent anonymous bad actors from accessing powerful AI tools, proposals increasingly demand mandatory identity verification and age verification. Executive Order 14110 established "Know Your Customer" (KYC) requirements for Infrastructure-as-a-Service (IaaS) providers, mandating that cloud computing companies verify the identities of foreign persons accessing compute resources to train large AI models35. Similarly, age verification mandates are proposed to shield minors from harmful AI-generated content.
The Advocate's Strongest Argument: Proponents argue that immense computational power is inherently dual-use; just as one cannot purchase weapons-grade uranium anonymously, one should not be able to train potentially catastrophic biological AI models anonymously3. KYC regulations are presented as a basic biosecurity and national security prerequisite. The Risk: Historical precedent indicates that identity verification mandates rapidly expand from foreign targets to domestic populations, effectively ending anonymous access to high-tier computational resources and linking all subsequent intellectual inquiries to a verified state identity.

Query Logging and Search-Query Profiling

As users rely on AI for information retrieval, platform providers systematically log prompts and interactions. These logs form a comprehensive psychological and intellectual profile of the user.
The Advocate's Strongest Argument: AI developers argue that query logging is essential for continuous model training, safety auditing, and incident response. If a user is utilizing an LLM to plan a terrorist attack, logging provides the necessary forensic trail to identify the threat and patch the vulnerability in the model37. The Risk: In the absence of strict privacy legislation, these logs become honey-pots for law enforcement. Through third-party subpoenas or National Security Letters, the state can easily access an individual's most private cognitive explorations, replicating the vulnerabilities exposed in the 2000s34.

Centralized AI Moderation and Content Classification

To prevent algorithmic propaganda, systemic misinformation, or the facilitation of crimes, centralized AI platforms employ automated moderation and content classification systems that scan inputs and outputs for prohibited concepts.
The Advocate's Strongest Argument: Existential risk theorists and safety advocates point out that unrestricted AGI could destabilize entire geopolitical realities through individually tailored propaganda or cyber-attacks3. Centralized moderation is viewed as the only mechanism capable of matching the scale and speed of machine-generated harm. The Risk: This creates a choke-point for human knowledge. Entrusting a handful of technology conglomerates to classify what constitutes "safe" or "truthful" information leads to the privatization of censorship, heavily influenced by state pressure.

Model Refusal Systems

To prevent AI systems from outputting harmful content (e.g., instructions for synthesizing pathogens), developers heavily fine-tune models to refuse specific prompts5.
The Advocate's Strongest Argument: Dual-use AI models possess the knowledge to dramatically lower the barrier to entry for biological terrorism5. Refusal systems are the final barrier preventing a lone actor from accessing world-ending capabilities. The Risk: Measuring refusal calibration is highly complex, and "false positives"—where a model refuses a benign or scientifically valid query—are rampant5. This creates an automated system of prior restraint, where a centralized AI decides what information a user is permitted to access, often acting opaquely and without due process or transparency.

Behavioral Risk Scoring and Automated Threat Detection

AI systems are increasingly deployed to scan vast datasets—including AI prompt logs—to detect patterns, flag suspicious activities, and score users based on behavioral risk6.
The Advocate's Strongest Argument: The speed of AI-driven cyber warfare or biological attacks outpaces human response times. Automated threat detection is necessary to preemptively neutralize threats before they materialize into kinetic harm3. The Risk: When applied to domestic populations, this enables predictive policing and automated behavioral risk scoring. Studies show that these algorithmic governance systems frequently perpetuate systemic discrimination, isolating human agency and reinforcing socio-economic biases while violating the presumption of innocence41.

Censorship Mandates and Platform-Government Coordination

Because AI models are treated as "black boxes" where traditional regulatory compliance (i.e., "code is law") cannot be explicitly verified, regulators lean toward outcome-based mandates37. Governments increasingly pressure centralized AI platforms to self-censor and report "risky" behavior.
The Advocate's Strongest Argument: The traditional regulatory model—where an expert agency audits a physical design (like an airplane)—fails with deep learning models, whose behaviors emerge unpredictably from training37. Therefore, close, continuous coordination between state intelligence and platform developers is required to monitor anomalous systemic risks. The Risk: This establishes a seamless, unaccountable conduit between corporate content moderation and state intelligence gathering, creating a neo-feudal equilibrium where a vanishingly small class of infrastructure owners wields power comparable to pre-Enlightenment monarchs, immune to democratic oversight39.

Part III: Structural Comparison: Evidentiary Similarities and Differences

By evaluating the historical patterns against contemporary proposals, the evidence highlights a profound trajectory. While the technologies have shifted from deterministic communication networks to non-deterministic neural networks, the sociological and institutional reflexes remain identical.

Evidentiary Similarities

Both the post-9/11 surveillance expansion and emerging AI governance frameworks rely on identical structural mechanisms to consolidate control.

Evidentiary Differences

Despite the similarities, crucial differences exist that make AI governance potentially more resilient to reform than earlier surveillance systems.
First, the nature of the technology is fundamentally different. In the 2000s, telephony networks were deterministic; it was clear what a phone record was, and rules could be explicitly coded. Conversely, advanced neural networks are non-deterministic "black boxes"37. Because human programmers do not write explicit rules for LLMs (i.e., "code is no longer law"), regulators argue that traditional compliance auditing is impossible, leading to a justification for continuous, real-time monitoring of user inputs and outputs37.
Second, the centralization of power in the AI era is heavily privatized. The mass surveillance of the 2000s was primarily a state-run endeavor that coerced the private sector. AI governance represents a "regulatory capture" dynamic39. Only the largest technology conglomerates possess the capital and compute resources to implement the massive automated safety, monitoring, and compliance architectures demanded by the state. This creates a symbiotic relationship where corporations welcome safety regulations to build a competitive moat, establishing an equilibrium where a handful of infrastructure owners wield unprecedented cognitive power over the populace39.

Part IV: Institutional Safeguards for AI Governance

The paramount question confronting contemporary policymakers is: What institutional mechanisms could prevent an AI-era “temporary safety measure” from becoming permanent infrastructure for monitoring lawful thought and inquiry?
The historical failures of the USA PATRIOT Act, the FISC, and the NSL regime reveal exactly which safeguards were missing in the early 2000s. To prevent the establishment of a permanent, cognitive surveillance state, specific institutional and technological mechanisms must be strictly codified into any AI regulatory framework.

1. Warrants and the Abrogation of the Third-Party Doctrine

The most glaring flaw in post-9/11 surveillance was the reliance on the Third-Party Doctrine, which allowed the government to access massive troves of data simply because it was held by a telecommunications company7. The Supreme Court's decision in Carpenter v. United States signaled a retreat from this doctrine, recognizing that historical cell-site location information provides a comprehensive chronicle of a user's physical presence that requires a warrant31.
This logic must be statutorily extended to AI interactions. Because an AI prompt log functions as a highly intimate "personal journal," exposing a user's innermost thoughts, legal vulnerabilities, and medical conditions, it must be granted absolute Fourth Amendment protection34. Handing data to an AI company to use its service cannot be construed as a voluntary waiver of privacy. Any AI regulatory framework must explicitly state that government access to an individual's AI interaction history requires a warrant supported by probable cause, issued by an Article III judge, completely invalidating administrative subpoenas like NSLs for cognitive data.

2. Technological Separation of Identity from Inquiry

The most profound failure of post-9/11 surveillance was the seamless linkage of identity to metadata. To prevent AI safety measures from becoming thought-policing tools, governance frameworks must mandate the technological separation of a user's identity from their generative AI inquiries.
While mandatory identity verification (KYC) regulations may be strictly justified for corporate entities purchasing massive compute clusters to train foundational models35, applying identity or age verification to end-users querying models is fundamentally incompatible with a free society. Governance must incentivize zero-knowledge architectures and cryptographic anonymization, ensuring that even if a platform logs a prompt for safety training, it cannot be mathematically linked to a specific citizen's identity without immense, targeted cryptographic effort.

3. Strict Minimization and Data Deletion

The bulk collection of telephony metadata relied on the indefinite retention of irrelevant data on the premise that it might be useful in the future9. To prevent AI platforms from becoming honeypots for intelligence agencies, strict data minimization and deletion mandates are required. Platforms must be legally prohibited from retaining user prompt histories beyond the ephemeral duration necessary to process the query, unless explicitly opted-into by the user for personal convenience. By mandating rapid data deletion, the state deprives itself of the ability to conduct retroactive, dragnet searches of civilian thought patterns.

4. Adversarial Proceedings and Reforming Judicial Review

The historical record of the Foreign Intelligence Surveillance Court demonstrates that a secret, ex parte court will inevitably expand executive power, lower standards of access, and create secret law13. When the USA FREEDOM Act introduced the amicus curiae provision (50 U.S.C. § 1803), it acknowledged the necessity of an adversarial process in secret courts44.
Any judicial or regulatory body tasked with overseeing AI safety, approving broad algorithmic monitoring, or demanding access to AI user data must feature a permanent, fully cleared, institutionalized adversarial advocate. This advocate's sole mandate must be the protection of privacy, civil liberties, and free expression, actively contesting the government's interpretations of "safety" and "relevance"16. Furthermore, the legal interpretations and decisions of this body must be subject to mandatory declassification and public release, permanently banning the creation of "secret law" regarding what citizens are allowed to ask a computer.

5. Transparency, Auditability, and Appeal Rights

The abuse of National Security Letters was facilitated entirely by permanent gag orders that prevented recipients from challenging the government or notifying the public17. AI governance must statutorily ban indefinite non-disclosure agreements for surveillance requests, ensuring the public is aware of how often and why the government seeks AI records.
Furthermore, as AI model "refusal systems" function as automated censors, the calibration of these models must be heavily audited for false positives5. Because models regularly refuse benign queries under the guise of biosecurity or safety, users must have transparent appeal rights. The heuristics determining "safety" must be open to independent, academic audit, rather than shielded as proprietary trade secrets, preventing the silent, algorithmic suppression of lawful information seeking5.

6. Decentralization of AI Infrastructure

The concentration of AI capabilities in a few massive data centers mirrors the centralization of telecommunications in the 2000s, making it incredibly easy for the government to implement chokepoints. To resist a neo-feudal surveillance state, regulatory frameworks must explicitly protect and subsidize decentralized, open-source AI models39. Ensuring that highly capable models can be run locally on edge devices removes the third-party intermediary entirely, technologically enforcing the Fourth Amendment by keeping the data strictly on the user's personal hardware.

7. Ironclad Sunset Clauses

History shows that emergency legislation tends to become permanent infrastructure. Any law granting exceptional powers to monitor AI compute, restrict open-source proliferation, or compel platform data-sharing must include strict, non-renewable sunset clauses. If the capability is truly necessary for species survival or national security, the government must be forced to publicly defend the program's efficacy and pass new legislation every few years. This prevents the silent normalization of exceptional powers that characterized the two decades following the Patriot Act.

Conclusion

The evolution of surveillance from the days of the USA PATRIOT Act to the era of Artificial Intelligence represents a profound shift from tracking human movement and association to mapping human cognition and intellectual inquiry. The historical record is unequivocal: when emergency conditions are used to justify broad, secretive, and poorly overseen surveillance powers, those powers will inevitably expand. Mission creep will set in, exceptional powers will be normalized, and the foundational liberties of the public will be chilled.
The strongest arguments for AI restriction—existential risk, dual-use biosecurity, and systemic epistemic harm—are not trivial. However, mitigating these risks does not require the sacrifice of the Fourth Amendment, nor does it necessitate the construction of a permanent, neo-feudal surveillance architecture controlled by a cartel of technology platforms and intelligence agencies. By embedding absolute warrant requirements, strict data minimization, adversarial judicial review, decentralization, and the technological separation of identity from inquiry into the bedrock of AI governance, society can navigate the perilous adolescence of artificial intelligence without extinguishing the light of intellectual freedom.

Works cited

1. Government Surveillance: Overview | Law | Research Starters, https://www.ebsco.com/research-starters/law/government-surveillance-overview
2. EFF Analysis Of The Provisions Of The USA PATRIOT Act, https://www.eff.org/deeplinks/2003/10/eff-analysis-provisions-usa-patriot-act
3. The Adolescence of Technology - Dario Amodei, https://darioamodei.com/essay/the-adolescence-of-technology
4. James Broughel - Department of Justice, https://www.justice.gov/atr/media/1362436/dl?inline
5. BioRefusalAudit: Auditing Biosecurity Refusal Depth Using ... - arXiv, https://arxiv.org/html/2605.30162v1
6. PRIVACY VS. SECURITY: LEGAL DILEMMAS IN SURVEILLANCE, http://www.ijlra.com/details/privacy-vs-security-legal-dilemmas-in-surveillance-laws-by-shashvat-tiwari
7. Fact Sheet: Section 215 of the USA PATRIOT Act - CSIS, https://www.csis.org/analysis/fact-sheet-section-215-usa-patriot-act
8. articles reading for terrorism: section 215 of the usa patriot act and, https://scholarship.law.nd.edu/cgi/viewcontent.cgi?article=1092&context=jleg
9. Privacy and Civil Liberties Oversight Board NSA Bulk Telephone, https://publicintelligence.net/pclob-report/
10. Should the USA Patriot Act Be Reauthorized? - FindLaw, https://supreme.findlaw.com/legal-commentary/should-the-usa-patriot-act-be-reauthorized.html
11. Remember When the Patriot Act Debate Was All About Library, https://www.propublica.org/article/remember-when-the-patriot-act-debate-was-about-library-records
12. The Role of Librarians in Challenges to the USA PATRIOT Act, https://scholarship.law.unc.edu/cgi/viewcontent.cgi?referer=&httpsredir=1&article=1034&context=ncjolt
13. The History and Future of Mass Metadata Surveillance, https://www.pogo.org/analyses/the-history-and-future-of-mass-metadata-surveillance
14. How the Phone Dragnet Slipped Through Cracks in Oversight, https://www.aclutx.org/news/how-phone-dragnet-slipped-through-cracks-oversight/
15. Hearing of the Senate Judiciary Committee on Strengthening, https://www.intelligence.gov/ic-on-the-record-database/results/testimony/hearing-of-the-senate-judiciary-committee-on-strengthening-privacy-rights-and-national-security-oversight-of-fisa-foreign-intelligence-surveillance-act-surveillance-programs
16. The Two Faces of the Foreign Intelligence Surveillance Court, https://www.repository.law.indiana.edu/cgi/viewcontent.cgi?article=11216&context=ilj
17. Abuse of Power: FBI's Secret Demands for Personal Information, https://www.aclusocal.org/news/abuse-power-fbis-secret-demands-personal-information-increase-under-patriot-act/
18. National Security Letters - epic.org, https://archive.epic.org/privacy/nsl/
19. National security letter - Wikipedia, https://en.wikipedia.org/wiki/National_security_letter
20. constitutional secrecy: aligning national security letter nondisclosure, https://scholarship.law.duke.edu/cgi/viewcontent.cgi?article=1383&context=dlj
21. - THE INSPECTOR GENERAL'S INDEPENDENT REPORT ON THE, https://www.govinfo.gov/content/pkg/CHRG-110hhrg34175/html/CHRG-110hhrg34175.htm
22. Parallel construction - Wikipedia, https://en.wikipedia.org/wiki/Parallel_construction
23. NSA/FISA Webcast Series - NACDL, https://www.nacdl.org/Content/NSA-FISAWebcastSeries
24. DEA and NSA Team Up to Share Intelligence, Leading to Secret, https://www.eff.org/deeplinks/2013/08/dea-and-nsa-team-intelligence-laundering
25. Report Alleges Police Use Secret Evidence Collected By Feds To, https://www.capradio.org/news/npr/story?storyid=578369420
26. Is the DEA Using NSA Warrantless Surveillance Data in Domestic, https://nccriminallaw.sog.unc.edu/2013/08/07/is-the-dea-using-nsa-warrantless-surveillance-data-in-domestic-drug-investigations/
27. DEA Receiving Tips from the NSA | Colorado Federal Criminal, https://www.mdavidlindsey.com/blog/dea-receiving-tips-from-the-nsa/
28. The Right to Be Left Alone: Privacy in a Rapidly Changing World, https://www.csis.org/analysis/right-be-left-alone-privacy-rapidly-changing-world
29. Law enforcement & reader privacy - PEN America, https://pen.org/press-release/law-enforcement-reader-privacy/
30. A Comparative Study of Surveillance Laws in Democratic Societies, https://wahacademia.com/index.php/Journal/article/download/233/188/692
31. Building on Carpenter: Six New Fourth Amendment Challenges, https://www.nacdl.org/Content/Building-on-Carpenter-Six-New-Fourth-Amendment-Cha
32. Electronic Surveillance and FISA | The Belfer Center for Science and, https://www.belfercenter.org/publication/electronic-surveillance-and-fisa
33. Release of Three Opinions Issued by the Foreign Intelligence, https://www.intelligence.gov/ic-on-the-record-database/declassified/release-of-three-opinions-issued-by-the-foreign-intelligence-surveillance-court
34. ChatGPT, Claude, and the Fourth Amendment: - Best Lawyers, https://www.bestlawyers.com/article/chatgpt-claude-and-the-fourth-amendment/8191
35. US Executive Order on Safe, Secure, and Trustworthy AI, https://www.longtermwiki.com/wiki/E366
36. Proposed rules may lead to increased compliance requirements, https://www.securitymagazine.com/articles/100509-proposed-rules-may-lead-to-increased-compliance-requirements-ahead
37. When code isn't law: rethinking regulation for artificial intelligence, https://academic.oup.com/policyandsociety/article/44/1/85/7684910
38. National Security Letters (NSLs) - Electronic Frontier Foundation, https://www.eff.org/issues/foia/07656JDB
39. Geopolitical Fragmentation, AI, and the Coming Neo-Feudal Order[v1], https://www.preprints.org/manuscript/202604.0074/v1
40. Why Refusal Rate Misranks Frontier LLMs on Biological Research, https://arxiv.org/pdf/2605.21545
41. The Legal Challenges of AI, Surveillance, and Digital Governance, https://assajournal.com/index.php/36/article/download/588/855/891
42. Strategic Implications of Openness in AI Development, https://ora.ox.ac.uk/objects/uuid:83ea712f-aba3-4176-957a-3bb4af0209d6/files/r9w032303n
43. UNITED STATES v. Project for Privacy and Surveillance, https://caselaw.findlaw.com/court/us-4th-circuit/117218523.html
44. Amici Curiae as Special Masters in the Shadow of FISA, https://scholarlycommons.law.cwsl.edu/cgi/viewcontent.cgi?article=1783&context=cwlr
45. Amici Curiae for FISC Announced - Lawfare, https://www.lawfaremedia.org/article/amici-curiae-fisc-announced