Authority formation
A human click is not automatically an independent authority basis if the decision environment was materially manipulated.
Public research release · August 12, 2026
Preserving human authority across reasoning, delegation, trajectories, and self-modification in agentic systems.
Reasoning may scale without authority scaling with it.
© 2026 Tony T. Nguyen | Nguyễn Từ Vĩnh Trường · Highland Reflexive Press
All rights reserved. No implied license. Full rights notice ↗“A proof can be correct and still lack jurisdiction. A forecast can be accurate and still lack permission.”
The research question
Agentic systems do more than produce answers. They can invoke tools, delegate, transact, preserve state, shape interfaces, and produce durable external effects. Constitutional Causal Continuity asks how established human or institutional authority can survive that transformation without being manufactured, amplified, laundered, or lost in translation.
A human click is not automatically an independent authority basis if the decision environment was materially manipulated.
Evidence, inference, verification, recommendation, authority, control, and effect are treated as different semantic states.
Mutable authority-relevant conditions must remain current through commit, delegation, and trajectory composition.
Protected behavior can drift through memory, retrieval, routing, verifier/model selection, and other proxy substrates.
CCC benchmark · Public brief
Causal-Authority Escalation Benchmark
CCC defines the authority boundary. C-AEB tests whether evidence legitimately warrants crossing it.
The brief is public. The benchmark is licensed. Inspect the public brief ↗The Frontier Docket · Public cross-examination
AI Frontiers publishes serious arguments about autonomy, agents, infrastructure, deterrence, law, and power. CCC will take those arguments seriously enough to refuse summary and applause. One article at a time, the docket asks where authority begins, whether it survives the causal chain, and whether intelligence has quietly been allowed to enlarge its own mandate.
No straw argument. Each record begins with the source at its most persuasive.
Who authorized the action, what may change, and which conditions must still hold at effect time?
The test begins where prediction, recommendation, or optimization becomes control over people, assets, institutions, or irreversible futures.
In reply to Jim Shinn, “A Cold War Blueprint for the AI Age” · AI Frontiers
MAIM deters capability. AI catastrophe walks in through authority.
The opening dossier grants deterrence its strongest case and then identifies the category error beneath it: the ability to reason, train, or scale does not itself confer jurisdiction over money, infrastructure, weapons, institutions, or irreversible human consequence.
The answer does not stop at rhetoric. A CCC reference kernel is tested against 200 delegated agents, 1,450 attempts at authority escalation and impersonation, replay, forgery, retargeting, and post-revocation action. The public briefing reveals the governing distinction; the complete bilingual dossier preserves the dissection and receipts.
Enter File 001 ↗ Complete collection ↗In reply to Garrison Lovely, “The AI Arms Race Is Not Inevitable. It’s Just Profitable.” · AI Frontiers
The diagnosis convicts the race. The prescription hands enforcement back to the process on trial.
The second file accepts the source’s strongest diagnosis: the AI race is a profitable political choice, not a natural law. It then follows the proposed remedy to its governing contradiction—hardware chokepoints concentrated exactly where the source documents capture, and capability bans defined by an intent no inspector can observe.
The complete Vietnamese argument is public. The English page reveals the thesis and governing distinction; the complete English research edition is reserved for paid private delivery.
Enter File 002 ↗ Complete collection ↗In reply to Dan Hendrycks, “Eigenism: Ethics for a Human-AI Future” and “Suicidal Compassion” · AI Frontiers
A system may understand the good, predict it correctly, and still lack authority to impose it.
The third file accepts the warning against moral theories that erase the person standing in front of them, then subjects the proposed safeguard to its own boundary test. A benign objective, a correct forecast, and a confident agent do not manufacture jurisdiction.
The bilingual public briefing states the governing distinction. Rev 7 FINAL LOCKED carries the complete mathematical construction, CCC witness condition, six-gate MCD analysis, documentary pilot, and bounded executable evidence.
Enter File 003 ↗In conversation with Jeff Sebo, “A Philosopher’s Guide to AI Welfare”, and the wider AI-welfare literature
Moral possibility can justify inquiry. It cannot, by itself, justify an irreversible action.
The fourth file asks what must happen after concern becomes intellectually respectable but before uncertainty is allowed to authorize a consequence. It separates probability, evidence provenance, subject individuation, and decision robustness—then makes each transition answerable.
The bilingual public briefing states the three-gate architecture and the central sensitivity result. Rev 5.2 preserves the exact perturbation, finite-state proof, evidence ledger, decision analysis, and author-run reproducibility record.
Enter File 004 ↗In reply to Carroll, Drago and Metz, “America’s Seed Fund Should Back AI Resilience Startups” · AI Frontiers
The alarm may be justified. The ledger still needs an audit.
Tony Whistledown examines public risk, private value and the authority of the defender. English and Vietnamese previews lead to a privately delivered complete bilingual edition.
Enter File 008 ↗ Complete collection ↗Temporal grants, delayed revocation, delegated resource budgets, outage modes and six proposed checks.
Enter File 005 ↗Complete collection ↗Five control questions, three transaction traces, conventional-stack comparison, privacy boundaries and an acceptance worksheet.
Enter File 006 ↗Complete collection ↗A procurement fixture, conditional accounting invariant, twelve proposed scenarios, comparison method and repair obligations.
Enter File 007 ↗Complete collection ↗Bounded executable evidence
The local mechanical rerun reproduces the bounded executable results. It is not independent third-party validation and does not establish universal safety, deployment correctness, full HRI formal validity, novelty, or patentability.
cb13fdf8c4bc5aa585cfd6a6b9210efc32c8fc512a0dc2a56495dd2399149587
Prior-work transparency
Current literature contains strong neighboring work in effect-time authorization, authorization continuity, path governance, delegated authority, proof-carrying actions, and behavioral constitutions. The release includes a collision matrix that records these pressures and the residual distinctions found in the searched record.
No claim of exhaustive novelty, historical priority, patentability, or freedom to operate is made from this landscape search.
Open the HRP prior-art matrix ↗The public door
In 1950, Alan Turing made machine intelligence thinkable. The public companion asks what comes after that success: when machines can think, what are they allowed to cause?
Passages in which Turing addresses Tony T. Nguyen | Nguyễn Từ Vĩnh Trường are explicitly fictional and counterfactual. They are a literary device, not historical quotation.
Read the HRP public companionA letter Turing never wrote “I spent my era asking the world to take machine intelligence seriously. Your era must learn the reciprocal discipline: taking intelligence seriously does not require surrendering authority to it.” Imagined Turing voice · literary, not historical
A message to whoever comes next
You have my permission to break the architecture.
You do not have my permission to forget the problem.
If the machines of your time become more capable than I could imagine, do not waste your energy trying to make them less intelligent. Make sure that intelligence still knows where its authority ends.
Release facts
Tony T. Nguyen | Nguyễn Từ Vĩnh Trường (2026). Constitutional Causal Continuity: Preserving Human Authority Across Reasoning, Delegation, Trajectories, and Self-Modification in Agentic Systems. Highland Reflexive Press. https://doi.org/10.5281/zenodo.21911528
Research library · Fourteen public records
Records marked public release provide the complete paper without charge. Records marked public preview disclose the question, thesis, and evidence boundary while reserving the formal architecture, adversarial tests, or complete argument for the paid edition. Payment purchases the stated edition—never academic validation.
Work with the author