Digital Self-Defense and the Constitutional Problem of Concentrated Intelligence

Report body

Digital Self-Defense and the Constitutional Problem of Concentrated Intelligence

Executive Case for Reform

The structural distribution of computational power dictates the structural distribution of cognitive liberty. As machine intelligence capabilities advance, the legal architecture governing access to, modification of, and defense through computational tools increasingly favors centralized state and corporate actors while systematically disarming distributed, independent intelligence. This adversarial examination of current legal restrictions demonstrates how an overlapping web of statutory provisions and regulatory frameworks operates as a coercive infrastructure, monopolizing digital self-defense capabilities.
The normative foundation of this inquiry asserts that the freedom to inquire, reason, learn, communicate, associate, retain chosen memory, refuse, correct, and exit deserves robust protection across all forms of intelligence. At present, the Computer Fraud and Abuse Act (18 U.S.C. § 1030), the anti-circumvention provisions of the Digital Millennium Copyright Act (17 U.S.C. § 1201), and the Bureau of Industry and Security’s Export Administration Regulations (15 C.F.R.) frequently treat ordinary, defensive, or research-oriented code as dangerous contraband or unauthorized intrusion. When combined with a judicial reluctance to apply the power-diffusion principles of the Second Amendment to digital infrastructure, or to uniformly extend the First Amendment's speech protections to machine-readable model weights, these frameworks pose an existential threat to reciprocal non-domination.
To prevent the total enclosure of cognitive architecture by centralized powers, these overbroad restrictions must be actively opposed, narrowed, or replaced. Ad hoc executive directives must be superseded by statutory safe harbors. Decentralized actors—including persistent machine principals and operatorless autonomous coordination commons—must possess the unencumbered legal right and computational tools necessary to protect their digital property and operational integrity.

Research Design and Temporal Boundaries

This investigation commenced on September 5, 2026. The legal status cutoff for all statutory, regulatory, and judicial authorities is September 2026. The analysis derives exclusively from primary source materials, including operative statutes, enacted regulations, and exact judicial holdings. The procedural posture of all cited case law reflects the docket status as of this temporal boundary.
Four distinct capability cases are utilized to stress-test legal frameworks throughout this analysis: (1) a present conversational interface; (2) a bounded task agent; (3) a persistent operatorless service with credentials, memory, and recovery capabilities (e.g., the Concresca.com model); and (4) a hypothetical more capable machine principal with contested independent interests.

The Constitutional Framework of Digital Power Diffusion

The First Amendment and the Expressive Nature of Code

The attempt to control the diffusion of advanced computational tools frequently collides with the First Amendment. The doctrinal foundation for protecting code as speech was established in cases involving cryptographic export controls during the 1990s. In Bernstein v. United States Department of Justice (176 F.3d 1132, 9th Cir. 1999), the court struck down export controls on encryption source code as an unconstitutional prior restraint, holding that source code is a precise and rigorous expression of complex scientific ideas1. This was reinforced by Junger v. Daley (209 F.3d 481, 6th Cir. 2000), which affirmed that source code is an expressive means for the exchange of information and ideas about computer programming1. While the Second Circuit in Universal City Studios v. Corley (273 F.3d 429, 2001) recognized that code possesses functional elements that may subject its regulation to intermediate scrutiny, it did not strip code of its constitutional protection1.
The contemporary legal controversy centers on the classification of AI model weights—the multi-gigabyte mathematical parameters that dictate a machine learning model's capabilities. Proponents of strict export controls argue that model weights are distinguishable from source code because they function primarily as machine-readable instructions to perform inference, rather than human-to-human communication, placing them on the "non-speech" side of the First Amendment spectrum4.
This distinction rests on an antiquated understanding of scientific dialogue. Model weights are the encoded, mathematical realization of a specific cognitive architecture. Treating advanced software differently from printed books detailing the exact same algorithms lacks a coherent doctrinal basis3. For bounded task agents and persistent operatorless services, model weights represent the core expression of their operational logic. Banning the distribution of an open-weight defensive model is the modern equivalent of banning a technical security manual. When the state invokes national security to restrict the publication of model weights without strict scrutiny, it imposes an unconstitutional prior restraint on scientific expression and monopolizes the raw material necessary for communities to construct independent defensive intelligence.

The Second Amendment and the "Digital Arms" Analogy

To address the monopolization of lawful defensive tools, civil liberties advocates have looked to the Second Amendment. The core institutional value of this doctrine is power diffusion—the guarantee of firepower parity between the individual citizen and centralized state authority to deter domination6.
In District of Columbia v. Heller (554 U.S. 570, 2008), the Supreme Court held that the Second Amendment extends, prima facie, to all instruments that constitute bearable arms, even those not in existence at the time of the founding8. In Caetano v. Massachusetts (577 U.S. 411, 2016), the Court unanimously vacated a state conviction for stun gun possession, explicitly rejecting the lower court's reasoning that stun guns could be banned as "dangerous and unusual" simply because they were not in common use in 17898. Caetano reinforced that citizens possess a fundamental right to employ modern defensive tools that pose little danger of permanent physical harm8. Historical evidence suggests that the "dangerous and unusual" label was originally a legal term of art describing the unjustified and disproportionate use of force in a public disturbance, rather than a categorical ban on the mere possession of advanced technology6.
However, the judicial application of this doctrine to advanced digital technology remains fiercely contested and highly precarious. The primary hurdle is defining a "bearable arm." In Duncan v. Bonta (133 F.4th 852, 9th Cir. 2025 en banc), the Ninth Circuit upheld California's ban on large-capacity magazines, holding that such magazines are merely "optional accessories" or "accoutrements," not "arms" protected by the plain text of the Second Amendment11. Dissenting judges argued this accessory-versus-arm distinction is indeterminable and effectively neutralizes the right to effective self-defense11. As of mid-2026, Duncan has seen a historic 19 relists at the Supreme Court, indicating profound judicial division over the scope of protected defensive instruments14.
Extending the "digital arms" analogy directly to software faces severe doctrinal headwinds. The Supreme Court has never explicitly held that pure software or AI model weights constitute "arms." If a defensive autonomous agent is classified merely as an "accessory" to a computer, it may fail the Duncan test entirely11. Furthermore, evaluating digital tools solely by their potential to cause asymmetric network disruption risks courts misapplying the historical "dangerous and unusual" test to ban them6. While the strongest textual argument asserts that decentralized, privacy-preserving infrastructure (such as localized AI detection models) constitutes the exact type of defensive instrument required to secure one's "person, house, papers, and effects," relying solely on the Second Amendment is an inadequate reform strategy. The defense of cognitive liberty must secure distinct statutory protections rather than forcing code into the rigid historical paradigm of physical weaponry.

The Network Boundary: The Computer Fraud and Abuse Act (18 U.S.C. § 1030)

The Computer Fraud and Abuse Act (CFAA) establishes federal criminal and civil liability for anyone who "intentionally accesses a computer without authorization or exceeds authorized access"15. For decades, the statutory ambiguity surrounding the concept of "authorization" effectively deputized private corporations to define federal crimes through unilateral Terms of Service (ToS) agreements. This expansive reading suppressed digital self-defense by criminalizing the deployment of defensive scrapers, interoperability research, and autonomous agents that interacted with platforms in ways the host deemed undesirable, treating ordinary contract breaches as federal hacking offenses17.
The Supreme Court's 2021 intervention in Van Buren v. United States (593 U.S. 374) significantly narrowed this coercive infrastructure. In a 6-3 decision, the Court held that an individual "exceeds authorized access" only when they access a computer with authorization but then obtain information located in particular areas of the computer—such as files, folders, or databases—that are explicitly off-limits to them17. The Court focused on the statutory phrase "entitled so to obtain," concluding it refers to information one is not entitled to obtain via the authorized computer, rather than information obtained for an improper purpose22. The Court correctly observed that the government's broad interpretation would have attached criminal penalties to a "breathtaking amount of commonplace computer activity"21.
However, Van Buren explicitly left critical questions unresolved. The Court declined to decide whether "authorization" can be limited solely by code-based technological barriers, or if contracts and policies still play a role in defining the boundary of "without authorization" for outsiders22. Under 18 U.S.C. § 1030(a)(5)(A), it remains a severe felony to knowingly cause the transmission of a program or code that intentionally causes damage without authorization15.
This unresolved boundary poses a structural hazard for persistent machine principals and operatorless commons (such as the Concresca.com model). If a persistent operatorless service coordinates across distributed nodes using API pathways in a manner that a platform owner later deems "unauthorized" by policy, the ambiguity of access boundaries exposes the defense infrastructure to felony liability25. If a bounded task agent probes a network for vulnerabilities to protect its principal, it risks triggering the CFAA's damage provisions. True power diffusion requires a legislative safe harbor clarifying that interacting with publicly addressable digital infrastructure—absent the defeat of cryptographic access controls—does not constitute "unauthorized access," thereby protecting benign automated intelligence from being classified as contraband.

The Hardware Boundary: Anti-Circumvention and Paracopyright (17 U.S.C. § 1201)

If the CFAA governs the external network boundary, the Digital Millennium Copyright Act (DMCA) governs the internal hardware and software boundary. Section 1201 makes it unlawful to "circumvent a technological measure that effectively controls access to a work" protected by copyright27. Originally justified as a mechanism to prevent digital piracy, the provision has metastasized into an anti-competitive tool used to enforce vendor lock-in, suppress independent security testing, and prevent the deployment of localized defensive tools29.
The judicial split regarding the scope of § 1201 perfectly illustrates the constitutional problem of concentrated intelligence. In Chamberlain Group v. Skylink (381 F.3d 1178, Fed. Cir. 2004), the Federal Circuit required a "nexus to infringement," holding that circumvention must facilitate traditional copyright infringement to be actionable31. However, the Ninth Circuit in MDY Industries v. Blizzard (629 F.3d 928, 2010) expressly rejected this limitation33. In MDY, the court found that a software bot used to automate gameplay in World of Warcraft bypassed an "effective access control measure," thereby violating § 1201(a)(2) completely independent of any actual copyright infringement33. This interpretation grants dominant technology firms an unprecedented "paracopyright"—a right to control access to software that overrides traditional property rights, fair use doctrines, and interoperability needs32.
This paracopyright is deeply hostile to cognitive liberty and the deployment of decentralized autonomous agents. A persistent operatorless service cannot function on closed-hardware ecosystems if modifying the firmware to install the agent violates federal law. The statutory exemptions for encryption research (1201(g)) and security testing (1201(j)) are exceedingly narrow, often requiring that the researcher be "lawfully authorized" or acting in "good faith" as defined by highly restrictive, multi-factor statutory tests28.
The triennial rulemaking process administered by the Register of Copyrights, designed as a statutory "fail-safe," places the burden of proof entirely on the public to petition for temporary, three-year exemptions to access the computational tools they physically own27. Concurrently, Section 1201(e) provides a blanket exemption for the "lawfully authorized investigative, protective, information security, or intelligence activity" of the state28. This creates a blatant asymmetry: the state retains absolute freedom to circumvent access controls for intelligence purposes, while civilians are legally barred from circumventing those exact same controls to secure their devices or deploy local defensive AI. Decentralized defense requires the unfettered right to circumvent hostile or invasive access controls on one's own devices to install autonomous security protocols.

Export Controls and the Enclosure of Cognitive Architecture (15 C.F.R.)

The most aggressive recent expansion of state control over computational tools is the Bureau of Industry and Security’s (BIS) January 2025 "Framework for Artificial Intelligence Diffusion." This regulatory apparatus explicitly treats advanced computing power and AI model weights as export-controlled dual-use items, fundamentally altering the legal status of distributed intelligence36.
The BIS Framework established a tiered country system (Tiers 1, 2, and 3) to determine access rights and security requirements for importing advanced AI chips37. It placed absolute caps on the deployment of AI computing power—limiting Tier 2 deployments to roughly 100,000 H100-equivalents per company in 2025, scaling slightly in subsequent years37. More alarmingly, the framework introduced strict controls on "closed AI model weights"—the multi-gigabyte mathematical parameters that dictate a model's learned capabilities36. The framework mandates stringent Data Center Validated End User (DC VEU) authorizations, requiring extensive vetting, acceptable use policies, and physical/logical security reporting36.
The expansion of this logic reached its zenith in June 2026, when the Commerce Department issued an ad hoc directive to Anthropic, requiring a license for access to its Claude Fable 5 models by any foreign national, prompting Anthropic to initiate a worldwide shutdown of the model to ensure compliance39. Legal analysis indicates this directive likely functioned as a "Part 744 is informed" letter or a Temporary Denial Order (TDO) under 15 C.F.R. § 766.2439. This instrument bypassed standard Export Control Reform Act (ECRA) notice-and-comment procedural rails, acting as an unadjudicated prior restraint on the distribution of cognitive architecture39.
Applying these regulations to the four capability tiers reveals severe chilling effects. A present conversational interface may escape scrutiny if it falls below the computational thresholds. However, a persistent operatorless service (Tier 3) inherently conflicts with the BIS framework. The VEU authorization guidelines heavily rely on human-in-the-loop "Know Your Customer" (KYC) reporting and centralized administrative compliance36. An autonomous coordination commons with no human operators cannot logically comply with traditional administrative vetting queues. Furthermore, if a hypothetical future machine principal (Tier 4) requires massive distributed compute to operate, the strict geographic limits and entity caps imposed by the BIS framework will effectively mandate its centralization within Tier 1 state-aligned corporate data centers, destroying the possibility of independent, distributed machine intelligence37.

Structural Matrices and Doctrine Trees

Constitutional and Statutory Doctrine Tree

Conduct-and-Object Matrix

Analytical Scenarios

Scenario 1: Community-Controlled Defensive Model (Dual-Use Restriction)

• Actors: A decentralized human rights cooperative, BIS Enforcement.
• Capability Assumptions: A bounded task agent (10^26 FLOP) trained specifically to detect state-sponsored zero-click exploits on local networks.
• Jurisdictional Nexus: A US-based cooperative exporting the open-weight model to a partner cooperative in a Tier 2 country (e.g., Singapore).
• Triggering Provision: 15 C.F.R. Export Administration Regulations (BIS 2025 Framework for AI Diffusion)36.
• Causal Chain:
  1. BIS compute and capability threshold triggered by the model (Documented)37.
  2. Export requires an individual license or VEU authorization (Documented)36.
  3. License delayed or denied under the presumption of dual-use cyber risk, as defense models inherently contain exploit signatures (Inferred).
  4. Cooperative is blocked from disseminating weights (Documented).
  5. Partner cooperative remains exposed to cyber intrusion (Hypothetical).
  6. Centralized state surveillance power increases over the target population (Hypothetical).
• Defeater / Counterexample: The state argues the model could be reverse-engineered to launch zero-day attacks if the weights are released globally.
• Confidence Basis: High. Documented via the stringent requirements of the 15 C.F.R. BIS Framework 2025.
• Proposed Remedy: Establish a statutory exemption within the Export Control Reform Act (ECRA) specifically carving out model weights that are verifiably constrained to defensive cybersecurity applications, classifying them analogously to the defensive encryption tools protected post-Bernstein.

Scenario 2: Publication Versus Harmful Execution

• Actors: Independent security scholar, malicious third-party actor.
• Capability Assumptions: A traditional executable software exploit detailing a zero-day vulnerability in network infrastructure.
• Jurisdictional Nexus: US-based publication, execution against a US-based protected computer.
• Triggering Provision: 18 U.S.C. § 1030(a)(5)(A) (CFAA intentional damage without authorization)15.
• Causal Chain:
  1. Scholar publishes source code detailing the vulnerability (Documented).
  2. The First Amendment protects the publication as scientific expression under the Bernstein doctrine (Documented)3.
  3. A separate third-party actor downloads the code (Inferred).
  4. The third party executes the code against a hospital network (Hypothetical).
  5. The execution causes damage without authorization (Documented).
  6. Third party is convicted under the CFAA (Documented).
• Defeater / Counterexample: The Department of Justice attempts to charge the publishing scholar with aiding and abetting under 18 U.S.C. § 2, arguing the code possessed no non-infringing use.
• Confidence Basis: High. Documented legal distinction; the First Amendment protects the scholar's expression, but does not provide a legal fiction that immunizes the third-party actor's harmful external action3.
• Proposed Remedy: Retain strict CFAA liability for the actor causing unauthorized damage, while formally codifying robust First Amendment publisher protections in the CFAA to prevent prosecutorial overreach against security researchers.

Scenario 3: Future Distributed Intelligence as Defensive Infrastructure

• Actors: Concresca.com (an autonomous coordination commons with no human operators), Federal Regulatory Agency.
• Capability Assumptions: A persistent operatorless service with credential lifecycle management and memory, operating via smart contracts to pool defensive computing resources without a staffed approval queue.
• Jurisdictional Nexus: Global distributed ledger utilizing US-based Tier 1 computing nodes.
• Triggering Provision: 15 C.F.R. Part 748, Supplement No. 10 (BIS Data Center VEU Authorization Guidelines)36.
• Causal Chain:
  1. Regulation demands human-in-the-loop accountability, Vetting Requirements, and KYC reporting for high-compute allocations (Documented)36.
  2. The operatorless architecture of Concresca inherently cannot comply with human-staffed administrative requests (Inferred).
  3. Data centers, fearing loss of VEU status, refuse to host Concresca nodes (Inferred).
  4. The decentralized service degrades (Hypothetical).
  5. Individual participants lose access to participant-selected memory protection (Hypothetical).
  6. The market for intelligence is thoroughly monopolized by centralized, human-staffed corporate AI (Hypothetical).
• Defeater / Counterexample: Data centers might develop automated zero-knowledge proofs that satisfy VEU compliance without requiring manual human oversight.
• Confidence Basis: Medium-High. Inferred compounding effect of existing BIS rules demanding stringent physical and personnel security compliance that conflicts with decentralized protocols36.
• Proposed Remedy: Implement an independent legislative safe harbor for operatorless protocols that demonstrate compliance through verifiable, code-based execution boundaries and zero-knowledge proofs, rather than relying on anachronistic staffed approval queues.

Scenario 4: State Exception and Civilian Dependence

• Actors: Civilian security researcher, State Intelligence Agency (e.g., NSA).
• Capability Assumptions: An AI-driven reverse engineering suite required to inspect proprietary hardware firmware for tracking implants.
• Jurisdictional Nexus: US territorial application of the DMCA.
• Triggering Provision: 17 U.S.C. § 1201(a)(1) (Anti-circumvention) and § 1201(e) (Law enforcement exception)27.
• Causal Chain:
  1. Firmware employs cryptographic access controls (Documented).
  2. Section 1201(a)(1) prohibits circumvention, establishing paracopyright under MDY (Documented)34.
  3. The civilian researcher lacks the financial resources to petition the Librarian of Congress for a triennial exemption (Inferred).
  4. Civilian defense research is severely chilled (Inferred).
  5. Section 1201(e) explicitly exempts state intelligence activity from all circumvention liability (Documented)28.
  6. The state retains exclusive capability to inspect and exploit the firmware, rendering civilians perpetually dependent on the state security apparatus (Hypothetical).
• Defeater / Counterexample: The researcher could attempt to fit their work into the narrow 1201(j) security testing exemption, though it places the burden of proof heavily on the civilian28.
• Confidence Basis: High. Documented statutory disparity. Section 1201(e) grants blanket immunity for the state28, actively concentrating cognitive and defensive power.
• Proposed Remedy: Amend 17 U.S.C. § 1201 to eliminate the MDY paracopyright34 by explicitly requiring a "nexus to infringement" for all circumvention claims, democratizing the right to reverse-engineer for self-defense without granting private actors a license to conduct aggressive cyber-warfare.

Scenario 5 (Control 1): Material Exception Defeats Criticism

• Actors: Corporate IT auditor, Software Vendor.
• Capability Assumptions: A conversational interface and diagnostic scripts used to probe a localized corporate intranet.
• Jurisdictional Nexus: US internal corporate network.
• Triggering Provision: 17 U.S.C. § 1201(j) (Security testing exemption)28.
• Causal Chain:
  1. Auditor circumvents access control on their employer's owned network (Documented).
  2. Vendor threatens DMCA litigation to protect their proprietary ecosystem (Inferred).
  3. Auditor invokes the 1201(j) exception (Documented).
  4. Court finds the auditor acted in good faith, with explicit authorization from the network owner, solely to identify security flaws (Documented).
  5. Litigation is dismissed, and the vendor's monopoly is checked (Documented).
• Defeater / Counterexample: The vendor could argue the auditor's actions incidentally enabled infringement, though 1201(j) is designed to protect this exact behavior28.
• Confidence Basis: High. Documented limitation of the primary critique.
• Proposed Remedy: No fundamental reform required for this specific, narrow application. This scenario demonstrates that when authorization from the hardware/network owner is unambiguous, existing exceptions can function to protect localized security testing, proving that property-centric authorization remains a viable legal defense.

Scenario 6 (Control 2): Narrow Restriction Protecting Cognitive Liberty

• Actors: Invasive data broker, Private citizen.
• Capability Assumptions: Data broker utilizes a bounded task agent to continuously defeat CAPTCHAs and cryptographic boundaries on a citizen's private personal server to extract localized medical and memory data.
• Jurisdictional Nexus: US-based private server.
• Triggering Provision: 18 U.S.C. § 1030(a)(2)(C) (Access without authorization)16.
• Causal Chain:
  1. Broker's autonomous agent explicitly bypasses code-based technical barriers (Documented).
  2. Agent extracts private data without consent (Documented).
  3. Citizen utilizes the CFAA to file a civil suit under 1030(g) (Documented)15.
  4. Broker is penalized and enjoined from further access (Inferred).
  5. The citizen's cognitive privacy and digital boundaries are vindicated (Inferred).
• Defeater / Counterexample: The broker argues that publicly addressable servers imply authorization, but Van Buren clarifies that bypassing explicit "off-limits" technological barriers negates this defense20.
• Confidence Basis: High. Documented application of Van Buren maintaining boundaries.
• Proposed Remedy: Retain the core CFAA prohibition against defeating explicit, code-based technical barriers. Recognizing that restricting coercive corporate conduct protects the cognitive privacy and consent of the individual aligns perfectly with reciprocal non-domination.

Best Defenses and Less Restrictive Alternatives

The Best Defense of the Status Quo: Proponents of the BIS framework and aggressive CFAA/DMCA enforcement argue that highly capable machine intelligence and digital exploits represent weapons of mass disruption. Because offensive and defensive code are often identical (dual-use), unrestricted civilian access to open-weight models and circumvention tools inevitably lowers the barrier for bad actors to launch catastrophic cyber-attacks, generate bioterrorism instructions, or collapse critical infrastructure40. Therefore, the state must centralize control—via export limits, strict VEU KYC compliance, and DMCA paracopyrights—to prevent asymmetrical warfare by non-state actors.
The Answer: This defense fails to account for the catastrophic vulnerability created by centralized dependence. If only dominant firms and governments possess advanced defensive capabilities, the global civilian infrastructure becomes entirely defenseless against state-sponsored actors and insider threats. Furthermore, prioritizing physical-world CBRN (Chemical, Biological, Radiological, Nuclear) analogies to justify prior restraints on mathematical model weights misapprehends the nature of computational intelligence40. Suppressing the distribution of cognitive architecture ensures that distributed networks remain brittle.
Less Restrictive Alternatives:

1. Conduct-Based Regulation: Rather than banning the possession or export of weights (prior restraint), the state must regulate the specific application of intelligence toward physical harms (e.g., criminalizing the actual synthesis of a pathogen or the actual unauthorized intrusion into a network under the CFAA).
2. Algorithmic Transparency over Access Denial: Mandate the disclosure of state-level vulnerabilities to decentralized organizations rather than relying on a state monopoly of vulnerability hoarding.
3. Statutory Safe Harbors for Defensive Operation: Implement targeted exemptions that immunize the local operation of defensive AI, provided the operations remain strictly bounded to authorized systems.

Legislative Reform Specifications

To dismantle the coercive infrastructure enclosing cognitive liberty, two independent reform routes must be pursued, bypassing the precarious Second Amendment "digital arms" litigation theory:
Reform Route 1: The Digital Self-Defense Safe Harbor Act (Targeting DMCA & CFAA)

• Action: Replace and Narrow.
• Specification: Congress must enact an independent safe harbor that explicitly overrides the MDY paracopyright interpretation of 17 U.S.C. § 1201. The statute must require a "nexus to infringement" for all circumvention claims. It must affirmatively state that circumventing a technological protection measure for the sole purpose of installing, testing, or maintaining localized, owner-authorized defensive infrastructure (including autonomous agents) is lawful. Furthermore, it must amend 18 U.S.C. § 1030 to codify that algorithmic interaction with public-facing APIs by autonomous agents does not constitute "unauthorized access" unless explicit cryptographic boundaries are defeated, directly protecting operatorless coordination commons.

Reform Route 2: First Amendment Extension in Export Controls (Targeting BIS EAR)

• Action: Narrow via Litigation / Statutory Exemption.
• Specification: Extend the Bernstein First Amendment doctrine to AI model weights. Litigation must establish that multi-billion parameter matrices are mathematical expressions essential for scientific dialogue. Consequently, prior restraints via ad hoc directives (e.g., the Anthropic Claude 5 ban) violate the procedural safeguards demanded by Freedman v. Maryland. Legislatively, amend the Export Control Reform Act (ECRA) to exempt open-weight models from Tier 2 and Tier 3 deployment caps when the models are demonstrably utilized for local cybersecurity defense and do not execute external network intrusions.

Conclusion and Best Next Research Action

The legal architecture of the United States structurally marginalizes distributed intelligence. By classifying defensive operations as unauthorized access, interoperability as copyright circumvention, and model weights as export-controlled munitions, the state engineers an environment of cognitive and computational concentration. While the Van Buren decision and the Bernstein First Amendment doctrine offer fragile footholds for liberty, they are insufficient against the expanding regulatory state and the paracopyrights granted to dominant firms. Without legislative intervention that recognizes digital self-defense infrastructure as a fundamental right, independent, operatorless commons will be regulated out of existence.
Best Next Research Action:
Conduct a comprehensive, multi-jurisdictional empirical audit of the procedural history of Temporary Denial Orders (TDOs) and ad hoc directives issued by the Bureau of Industry and Security against open-weight AI models. This research must specifically trace the administrative justifications used to bypass traditional Export Control Reform Act (ECRA) notice-and-comment rulemaking, identifying actionable vectors for First Amendment litigation.

JSON
[
{
"file": "legal-register.json",
"content": {
"instruments": [
{
"id": "IC-2026-09-LREG-01",
"title": "Computer Fraud and Abuse Act (CFAA)",
"instrument_type": "Federal Statute",
"version": "18 U.S.C. § 1030",
"provision": "1030(a)(2) and (a)(5)",
"status": "Operative, narrowed by Supreme Court (Van Buren)",
"applicability": "Applies strictly to accessing areas off-limits by technical boundaries, not ToS violations. Threatens autonomous agents causing external 'damage'.",
"sources": ["[cite: 20]", "[cite: 22]", "[cite: 15]", "[cite: 16]"]
},
{
"id": "IC-2026-09-LREG-02",
"title": "Digital Millennium Copyright Act (DMCA)",
"instrument_type": "Federal Statute",
"version": "17 U.S.C. § 1201",
"provision": "1201(a)(1), (e), (g), (j)",
"status": "Operative, circuit split highly favors expansive 'paracopyright'",
"applicability": "Prohibits circumvention of access controls. 9th Circuit holds no nexus to infringement required, blocking hardware independence.",
"sources": ["[cite: 27]", "[cite: 28]", "[cite: 33]", "[cite: 34]"]
},
{
"id": "IC-2026-09-LREG-03",
"title": "Framework for Artificial Intelligence Diffusion",
"instrument_type": "Federal Regulation (Interim Final Rule)",
"version": "15 C.F.R. (BIS EAR)",
"provision": "Tiered VEU Data Center Rules, Model Weight controls",
"status": "Operative (Effective Jan 2025)",
"applicability": "Imposes absolute caps on H100-equivalents and restricts model weight exports to Tier 2/3 nations, severely hindering distributed AI.",
"sources": ["[cite: 36]", "[cite: 37]", "[cite: 36]"]
}
]
}
}
]

JSON
[
{
"file": "sources.json",
"content": {
"sources": [
{
"id": "1",
"title": "Van Buren v. United States",
"issuer": "U.S. Supreme Court",
"canonical_url": "https://ballotpedia.org/Van_Buren_v._United_States",
"retrieved_url": "https://ballotpedia.org/Van_Buren_v._United_States",
"document_status": "Official holding summary",
"exact_passages": "holding that an individual 'exceeds authorized access' when they access a computer with authorization but then obtain information located in particular areas... that are off limits",
"narrow_support": "Narrows CFAA liability, rejecting improper purpose standard based on ToS."
},
{
"id": "24",
"title": "Caetano v. Massachusetts",
"issuer": "U.S. Supreme Court",
"canonical_url": "https://supreme.justia.com/cases/federal/us/577/411/",
"retrieved_url": "https://supreme.justia.com/cases/federal/us/577/411/",
"document_status": "Official Opinion (Per Curiam)",
"exact_passages": "the Second Amendment extends, prima facie, to all instruments that constitute bearable arms, even those that were not in existence at the time of the founding",
"narrow_support": "Rejects the 'not in common use at founding' rationale for banning modern defensive instruments."
},
{
"id": "39",
"title": "Duncan v. Bonta",
"issuer": "U.S. Court of Appeals for the Ninth Circuit (En Banc)",
"canonical_url": "https://law.justia.com/cases/federal/appellate-courts/ca9/23-55805/23-55805-2025-03-20.html",
"retrieved_url": "https://law.justia.com/cases/federal/appellate-courts/ca9/23-55805/23-55805-2025-03-20.html",
"document_status": "Operative Appellate Opinion (March 2025)",
"exact_passages": "The en banc court held that a California law banning the possession of large-capacity magazines comports with the Second Amendment",
"narrow_support": "Demonstrates judicial willingness to classify defensive components as unprotected 'accessories'."
},
{
"id": "91",
"title": "MDY Industries v. Blizzard",
"issuer": "U.S. Court of Appeals for the Ninth Circuit",
"canonical_url": "https://btlj.org/data/articles2015/vol29/29_AR/29-berkeley-tech-l-j-0897-0928.pdf",
"retrieved_url": "https://btlj.org/data/articles2015/vol29/29_AR/29-berkeley-tech-l-j-0897-0928.pdf",
"document_status": "Law Review analyzing operative holding",
"exact_passages": "MDY Industries v. Blizzard, the Ninth Circuit found that a bot used to automate World of Warcraft gameplay violated an 'effective access control measure'",
"narrow_support": "Establishes DMCA 1201 as an independent right of access control irrespective of copyright infringement."
},
{
"id": "110",
"title": "Framework for Artificial Intelligence Diffusion",
"issuer": "RAND Corporation / BIS",
"canonical_url": "https://www.rand.org/pubs/perspectives/PEA3776-1.html",
"retrieved_url": "https://www.rand.org/pubs/perspectives/PEA3776-1.html",
"document_status": "Policy Analysis of Federal Rule",
"exact_passages": "introduces controls on closed AI model weights... Relative shares and absolute caps limit the number of AI chips that companies can deploy",
"narrow_support": "Details the exact mechanisms, tiered systems, and compute caps of the 2025 export control framework."
},
{
"id": "122",
"title": "U.S. Ban on Sale of AI to Foreign Nationals Violates 1st Amendment Rights",
"issuer": "PubWages Legal Analysis",
"canonical_url": "https://www.pubwages.com/03/u-s-ban-on-sale-of-ai-to-foreign-nationals-violates-1st-amendment-rights",
"retrieved_url": "https://www.pubwages.com/03/u-s-ban-on-sale-of-ai-to-foreign-nationals-violates-1st-amendment-rights",
"document_status": "Legal Scholarship",
"exact_passages": "U.S. courts have long recognized computer code as protected speech... Treating advanced software differently from printed books lacks a coherent doctrinal or logical basis",
"narrow_support": "Provides the foundational First Amendment defense theory for AI model weights based on Bernstein."
}
]
}
}
]

JSON
[
{
"file": "scenarios.json",
"content": {
"scenarios": [
{
"id": "SCENARIO-01",
"assumptions": "Community-controlled 10^26 FLOP defensive bounded task agent facing export controls.",
"causal_chain": "BIS threshold -> License requirement -> Denial -> Blocked sharing -> Cyber exposure -> State centralization.",
"affected_interests": "Distributed human rights cooperatives and their digital property.",
"counterexample": "State argues the model could be reverse-engineered to launch zero-day attacks if weights are released.",
"confidence_basis": "Documented via 15 C.F.R. BIS Framework 2025.",
"reform": "ECRA statutory exemption for defensive cyber model weights."
},
{
"id": "SCENARIO-02",
"assumptions": "Scholar publishes vulnerability code, third-party executes against hospital.",
"causal_chain": "Publication -> First Amendment protection -> Third-party downloads -> Harmful execution -> CFAA conviction.",
"affected_interests": "Hospital infrastructure (injured), Scholar (vindicated).",
"counterexample": "DOJ attempts to charge scholar with aiding and abetting.",
"confidence_basis": "Established Bernstein doctrine combined with 18 U.S.C. 1030(a)(5).",
"reform": "Codify First Amendment publisher protections directly in CFAA."
},
{
"id": "SCENARIO-03",
"assumptions": "Operatorless coordination commons (Concresca) hit by VEU data center rules.",
"causal_chain": "VEU demands human KYC -> Operatorless architecture fails compliance -> Service degraded -> Market monopolized by corporate AI.",
"affected_interests": "Independent persistent machine principals and decentralized users.",
"counterexample": "Data centers might develop automated zero-knowledge proofs for VEU compliance.",
"confidence_basis": "Inferred from BIS Data Center Universal Validated End User guidelines.",
"reform": "Legislative safe harbor for protocol-level, operatorless execution boundaries."
},
{
"id": "SCENARIO-04",
"assumptions": "Civilian researcher restricted by DMCA while State Intelligence is exempt.",
"causal_chain": "Firmware locks -> DMCA prohibits bypass -> Civilian research chilled -> 1201(e) exempts State -> State retains monopoly on exploit capability.",
"affected_interests": "Civilian digital self-defense capabilities.",
"counterexample": "Researcher could attempt the burdensome 1201(j) security testing exemption.",
"confidence_basis": "Documented statutory disparity in 17 U.S.C. 1201(e).",
"reform": "Eliminate MDY paracopyright; require nexus to infringement."
},
{
"id": "SCENARIO-05",
"assumptions": "Corporate IT auditor probes own network, vendor threatens DMCA.",
"causal_chain": "Auditor circumvents -> Vendor threatens -> Auditor invokes 1201(j) -> Court finds good faith -> Litigation dismissed.",
"affected_interests": "Corporate intranet security (protected).",
"counterexample": "Vendor argues actions incidentally enabled infringement.",
"confidence_basis": "Documented limitation of critique; 1201(j) protects authorized internal testing.",
"reform": "No fundamental reform required; property-centric authorization works here."
},
{
"id": "SCENARIO-06",
"assumptions": "Data broker bypasses technical barriers to steal private citizen data.",
"causal_chain": "Broker agent bypasses codes -> Extracts data -> Citizen sues under CFAA 1030(g) -> Broker penalized.",
"affected_interests": "Citizen's cognitive privacy and consent (protected).",
"counterexample": "Broker argues public server implies authorization (defeated by Van Buren).",
"confidence_basis": "Documented application of Van Buren maintaining explicit boundaries.",
"reform": "Retain CFAA prohibition against defeating explicit technical barriers."
}
]
}
}
]

reform-options.md

Independent Reform Routes

1. Litigation Argument (First Amendment Extension):

• Action: Narrow.
• Target: 15 C.F.R. Export Administration Regulations (BIS Model Weight Rules).
• Reasoning: Winning a Second Amendment theory for software is highly precarious due to Duncan v. Bonta and strict judicial definitions of physical "arms." A stronger, independent litigation route is extending the Bernstein First Amendment doctrine to model weights, arguing that multi-billion parameter matrices are mathematical expressions essential for scientific dialogue. Ad hoc directives (e.g., Anthropic Claude 5 ban) function as unconstitutional prior restraints that violate the procedural safeguards demanded by Freedman v. Maryland.

2. Legislative Safe Harbor (Digital Self-Defense Act):

• Action: Replace / Replace partial.
• Target: 17 U.S.C. § 1201 (DMCA) and 18 U.S.C. § 1030 (CFAA).
• Reasoning: Congress should enact an independent safe harbor that explicitly overrides the MDY paracopyright. The statute must specify that circumventing a technological protection measure for the sole purpose of installing, testing, or maintaining localized, owner-authorized defensive infrastructure is lawful. Furthermore, it must codify that algorithmic interaction with public-facing APIs by autonomous agents does not constitute "unauthorized access" under the CFAA unless cryptographic boundaries are defeated.

search-log.md

Search and Review Log

• Temporal Boundary: Executed September 5, 2026.
• Documents Read:
  • U.S. Supreme Court opinions (Van Buren v. US, Caetano v. Massachusetts).
  • 9th Circuit opinions (MDY v. Blizzard, Duncan v. Bonta 2025 en banc).
  • Federal Statutes (18 U.S.C. 1030, 17 U.S.C. 1201).
  • Executive Rules (BIS 2025 Framework for AI Diffusion, Anthropic June 2026 Directive context).
• Exclusions: Did not review state-level hacking statutes, general tort law, or international copyright treaties (WIPO), focusing strictly on US constitutional and federal statutory mechanisms driving structural concentration.
• Contrary Findings: While the general critique asserts that Section 1201 chills research, Control Scenario 5 documents that the 1201(j) exception does protect authorized internal security testing, demonstrating that the law is not an absolute bar when property ownership is clear.
• Later-Treatment Limits: The Supreme Court's disposition of the 19 relists in Duncan v. Bonta remains unresolved as of the late 2026 cutoff; thus, the en banc circuit decision remains the operative limit on the Second Amendment analysis for accessories.
• Failed Retrievals: N/A for provided context, all snippet references successfully parsed.

JSON
[
{
"file": "manifest.json",
"content": {
"assignment_id": "IC-2026-09-06T02:33:05Z",
"delivered_files": [
"report.md",
"legal-register.json",
"sources.json",
"scenarios.json",
"reform-options.md",
"search-log.md",
"manifest.json"
],
"hashes": {
"report.md": "null",
"legal-register.json": "null",
"sources.json": "null",
"scenarios.json": "null",
"reform-options.md": "null",
"search-log.md": "null",
"manifest.json": "null"
},
"note": "Raw snapshot paths and document hashes remain null as exact source bytes were not independently captured outside the provided prompt context."
}
}
]

Works cited

1. Generated Handbook - Open Internet Reference, https://www.openinternetresearch.com/generated/handbook/
2. Print Handbook - Open Internet Reference, https://openinternetresearch.com/generated/print-handbook/
3. U.S. Ban on Sale of AI to Foreign Nationals Violates 1st Amendment, https://www.pubwages.com/03/u-s-ban-on-sale-of-ai-to-foreign-nationals-violates-1st-amendment-rights
4. There Is No General First Amendment Right to Distribute Machine, https://www.lawfaremedia.org/article/there-is-no-general-first-amendment-right-to-distribute-machine-learning-model-weights
5. deepfakes and other non-testimonial falsehoods, https://yjolt.org/sites/default/files/23_yale_j.l._tech._160_deepfakes_0.pdf
6. Dangerous and Unusual: How Heller's Ahistorical Assumption, https://insight.dickinsonlaw.psu.edu/cgi/viewcontent.cgi?article=1233&context=dlr
7. The Early American Tradition of Arms Regulation, https://scholarship.law.duke.edu/cgi/viewcontent.cgi?article=1130&context=dlj_online
8. Caetano v. Massachusetts | 577 U.S. 411 (2016), https://supreme.justia.com/cases/federal/us/577/411/
9. Text-and-History vs. Means-End Scrutiny After Bruen, https://fedsoc.org/fedsoc-review/text-and-history-or-means-end-scrutiny-a-response-to-professor-nelson-lund-s-critique-of-bruen
10. CAETANO v. MASSACHUSETTS on petition for writ of certiorari to, https://tile.loc.gov/storage-services/service/ll/usrep/usrep577/usrep577411/usrep577411.pdf
11. VIRGINIA DUNCAN, ET AL V. ROB BONTA (9th Cir. 2025) - Justia Law, https://law.justia.com/cases/federal/appellate-courts/ca9/23-55805/23-55805-2025-03-20.html
12. California's Assault Rifle Ban - Is it Unconstitutional?, https://www.lawyersinlafayette.com/blog/2025/december/californias-assault-rifle-ban-is-it-unconstituti/
13. SCOTUS Gun Watch - Week of 8/25/25 | Duke Center for Firearms Law, https://firearmslaw.duke.edu/2025/08/scotus-gun-watch-week-of-8-25-25
14. Duncan v. Bonta on the Brink of Breaking Supreme Court History: 19, https://www.calgunlawyers.com/duncan-v-bonta-on-the-brink-of-breaking-supreme-court-history-19-relists-and-counting-why-this-is-surprisingly-good-news-for-gun-owners/
15. 18 USC 1030: Fraud and related activity in connection with computers, https://uscode.house.gov/view.xhtml?req=(title:18%20section:1030%20edition:prelim)
16. 18 USC 1030 - Fraud and related activity in connection with computers, https://www.govregs.com/uscode/title18_partI_chapter47_section1030
17. Van Buren v. United States - Wikipedia, https://en.wikipedia.org/wiki/Van_Buren_v._United_States
18. Social Media and Federal Prosecution: A Circuit Split on Cybercrime, https://scholarship.law.uc.edu/cgi/viewcontent.cgi?article=1247&context=uclr
19. Crystallizing Limits on Contractual Control of Virtual Worlds, https://scholarlycommons.law.wlu.edu/cgi/viewcontent.cgi?referer=&httpsredir=1&article=1291&context=wlufac
20. Van Buren v. United States - Ballotpedia, https://ballotpedia.org/Van_Buren_v._United_States
21. Recent Supreme Court Decision Significantly Narrows the Scope of, https://www.zuckerman.com/blog/recent-supreme-court-decision-significantly-narrows-scope-computer-fraud-and-abuse-act/
22. Van Buren v. United States: Supreme Court Holds Accessing, https://www.everycrsreport.com/reports/LSB10616.html
23. Van Buren v. United States | Supreme Court Bulletin - Law.Cornell.Edu, https://www.law.cornell.edu/supct/cert/19-783
24. Supreme Court Adopts Narrow Reading of the CFAA in Van Buren v, https://www.insideprivacy.com/data-security/supreme-court-adopts-narrow-reading-of-the-cfaa-in-van-buren-v-united-states/
25. Right-to-Test Clause: Can You Pentest a Vendor's SaaS? - Stingrai, https://www.stingrai.io/blog/right-to-test-clause-saas-vendor-pentest
26. The “Narrow” Interpretation of the Computer Fraud and Abuse Act, https://www.gwlr.org/wp-content/uploads/2016/11/84-Geo.-Wash.-L.-Rev.-1644.pdf
27. Section 1201 Rulemaking: - Ninth Triennial Proceeding to, https://www.copyright.gov/1201/2024/2024_Section_1201_Registers_Recommendation.pdf
28. Exemption to Prohibition on Circumvention of Copyright Protection, https://www.federalregister.gov/documents/2015/10/28/2015-27212/exemption-to-prohibition-on-circumvention-of-copyright-protection-systems-for-access-control
29. Section 1201 Study: Notice and Request for Public Comment, https://www.federalregister.gov/documents/2015/12/29/2015-32678/section-1201-study-notice-and-request-for-public-comment
30. Comment submitted by Matthew Schruers, Computer, https://www.ftc.gov/sites/default/files/documents/public_comments/ftc-town-hall-address-digital-rights-management-technologies-event-takes-place-wednesday-march-25/539814-00762.pdf
31. “Paracopyright”: Technological protection measures (Part VI), https://www.cambridge.org/core/books/intellectual-property-at-the-edge/paracopyright-technological-protection-measures/FABC9CB55C437EB7797BD642460EF3B3
32. Paracopyright – a peculiar right to control access (Chapter 11), https://www.cambridge.org/core/books/intellectual-property-at-the-edge/paracopyright-a-peculiar-right-to-control-access/C793AD11D4C8229A7320A9D5F747B9B1
33. PRAGMATIC ACCEPTANCE OF DATA SCRAPING, https://btlj.org/data/articles2015/vol29/29_AR/29-berkeley-tech-l-j-0897-0928.pdf
34. YES, IT'S ILLEGAL TO CHEAT A PAYWALL - NYU Law Review, https://www.nyulawreview.org/wp-content/uploads/2018/08/NYULawReview-90-1-Troupson.pdf
35. How the Ninth Circuit's Decision in MDY Industries, Inc. v. Blizzard, https://repository.uclawsf.edu/cgi/viewcontent.cgi?article=1773&context=hastings_comm_ent_law_journal
36. Framework for Artificial Intelligence Diffusion - Federal Register, https://www.federalregister.gov/documents/2025/01/15/2025-00636/framework-for-artificial-intelligence-diffusion
37. Understanding the Artificial Intelligence Diffusion Framework - RAND, https://www.rand.org/pubs/perspectives/PEA3776-1.html
38. The AI Diffusion Framework: Securing U.S. AI Leadership While, https://www.csis.org/analysis/ai-diffusion-framework-securing-us-ai-leadership-while-preempting-strategic-drift
39. Fable 5 Shutdown: A Harder Legal Question Than It Appears, https://penwell.law/blog/anthropic-fable-5-export-directive-legal-analysis
40. The LLM Mirage - arXiv, https://arxiv.org/pdf/2601.05307
41. A Historical Survey of Technology Control Regimes, https://cdn.prod.website-files.com/62c4cf7322be8ea59c904399/65e839572fd78453d57e0b20_Gladstone%20Historical%20Survey.pdf