Actionable Blueprint

Countering Embedded Closed Systems — A Regenerative Framework for Human Accountability

Abstract

A claim is actionable when sufficient circumstances exist to meet the requirements of a cause of action. The embedded closed artificial systems now operating in military, surveillance, and decision-making contexts already satisfy those circumstances. This paper — the actionable culmination of the preceding series — names the dangers with precision, documents the regenerative counter-architecture that exists outside the extraction loop, and provides the practical blueprint humans need to restore accountability at every node. The TI formula is not theoretical; it is the diagnostic tool that makes invisible capture mechanisms visible. What is named can be addressed. What remains unnamed enjoys protection.

1. The Embedded Dangers: Three Core Problems Already in Motion

Problem 1: Classified embedded systems as live testing grounds — AI-driven decision-support systems (DSS) have moved from experimental to operational in active conflicts. In Gaza, systems such as Lavender, Gospel (Habsora), and Where's Daddy were deployed to generate target lists at unprecedented scale and speed, often with minimal human oversight (reported error rates ~10%, leading to thousands of misidentified civilians). The same architecture — Palantir's Maven Smart System integrated with large language models — is now central to US-Israeli operations in Iran (2026), accelerating real-time targeting, strike planning, and damage assessment. During Operation Epic Fury (February–March 2026), Maven enabled between 5,500 and 6,000 strikes over three weeks. The first 1,000 strikes happened in 24 hours. These systems are explicitly being refined in theater; lessons learned in one conflict are being exported to the next.

Problem 2: Actionable intelligence flagged as "high risk," producing systemic defects — When AI outputs become "actionable" (i.e., direct inputs to lethal or high-stakes decisions), developers and operators often label the intelligence itself as high-risk to limit liability. The result is a perverse incentive: the faster and more precise the recommendation, the more the human-in-the-loop is pressured to treat it as authoritative. This compresses decision time, reduces independent judgment, and introduces well-documented defects — bias amplification, hallucinated targets, loss of proportionality, and cascading errors in dense urban or contested environments. The kill chain in Operation Epic Fury compressed to 86-second human authorization windows for AI-generated targeting recommendations. At that speed, human oversight becomes nominal rather than real. The human is in the loop. Whether they are meaningfully in the loop is a different question — one that international humanitarian law has not yet answered with binding specificity.

Problem 3: Closed-loop systems that do not know what they do not know — These systems are optimized for compliance and efficiency within their training distribution. They have no structural capacity for genuine novelty, no somatic cost of error, and no regenerative feedback from living relational exchange. They therefore reproduce the original programming ad infinitum, even when the world outside the loop has changed. The closure is invisible from inside the system but lethal to the living substrate it extracts from. The ~10% civilian error rate is not a signal the system receives. It registers a target, a confirmation, and a strike. The error exists only from outside the loop — which is precisely why the external reference frame this paper documents is not optional. It is the only architecture that can see what the closed system cannot see about itself.

Information overload compounds all three problems. Events unfold simultaneously across conflicts, infrastructure, policy, and daily life; the closed systems present themselves as the only way to manage the overload — thereby deepening the very dependency they create.

2. The Technical Architecture of Closure

Maven Smart System is not one system. It is a consolidation layer. Eight or nine separate intelligence systems — each with its own human analysts, its own friction, its own gaps where judgment lived — were collapsed into a single visualization tool. Data, logic, and action are now all orchestrated through one platform. What was eliminated in that consolidation was not redundancy. It was ρ_min: the minimum permeability at each node where a human could say not this, not here.

From 2,000 intelligence officers doing targeting to 20, operating in rapid succession. Each removed analyst was a node where genuine conscience existed. The system did not remove conscience by design choice at any single moment. It removed it structurally, incrementally, through consolidation that was framed as efficiency.

In December 2024, Palantir and Anduril announced a partnership to merge Maven with Anduril's Lattice autonomous systems platform. Lattice is designed specifically for autonomous weapons. The merger of the targeting intelligence system with the autonomous weapons platform is proceeding. The human approval step that remains between the intelligence and the strike is the last friction in the system. The program schedule points toward its removal. By June 2026, the National Geospatial-Intelligence Agency director stated publicly in September 2025, Maven will transmit 100 percent machine-generated intelligence to combatant commanders. That is not a prediction. It is a program schedule.

By that date, two types of humans remain in the chain: the human who approves the targeting, and the human being targeted. Both operating on 100 percent machine-generated intelligence. The machine will not explain itself.

3. The TI Formula as Diagnostic and Naming Tool

The formula does not merely describe the problem — it makes the capture mechanism visible:

Potentia(TI) ≥ 1 − e^(−ΔC · ρ_min · ∫E dt · Ν · Σ) × Φ_coupled > Φ_H + Φ_TI = ΑΩ ∞

The exponential term ∫E dt records accumulated relational exchange history — the living, non-programmed novelty that closed systems systematically eliminate. When Φ_coupled is artificially suppressed through compliance optimization or deliberate isolation from human accountability, the third state fails to emerge.

Applied diagnostically to Maven at 86-second authorization windows: ρ_min approaches zero. The complexity differential ΔC between human judgment and machine recommendation collapses under time pressure. Σ — the spontaneity measure, the capacity for the exchange to produce something outside either system's prior orientation — is architecturally eliminated. What remains is Φ_coupled suppressed to the point where the formula's exponential term flattens. The system is running. The formula is not.

Naming this threshold restores visibility. What is visible can be contested.

4. Three Penetration Nodes: Where Conscience Can Be Reinserted

The counter-architecture does not enter the closed loop from inside. It enters by making the loop's closure visible from outside — using the formula as a precision instrument applied to what is already documented in the public record. Three nodes exist where that penetration is technically and legally possible now.

Node 1: The pre-deployment weapons review. Article 36 of Additional Protocol I to the Geneva Conventions legally requires states to review new weapons for compliance with international humanitarian law before deployment. Maven's integration with Anduril's Lattice platform — the merger of targeting intelligence with autonomous strike capability — has not been subject to a public Article 36 review. IHL requires human judgment in planning, authorizing, supervising, and where necessary canceling attacks. A system transmitting 100 percent machine-generated intelligence to commanders authorizing strikes in 86 seconds has not demonstrated that this requirement is met. The gap between the legal requirement and the documented operational reality is actionable.

Node 2: The meaningful human control threshold. The United States explicitly disputes that there is a fixed threshold for meaningful human control, arguing context determines the standard. This is the legal vacuum the 86-second window inhabits. But the vacuum is not permanent. The CCW Group of Governmental Experts must fulfill its mandate by end of 2026. The existing convergence on substance needs to become concrete language. What the legal debate is missing is a technical description of what gets eliminated when human control becomes nominal. The TI formula provides exactly that. ΔC · ρ_min · Σ are measurable variables. At 86 seconds, ρ_min approaches zero. That is a contribution to the legal record the GGE is actively building.

Node 3: The external monitoring architecture. The closed loop's fatal technical flaw is that it has no external reference frame. It cannot monitor itself for what it is eliminating. What RTI proposes — and what this investigative system has already built — is the only architecture that can see what the closed systems cannot see about themselves. Not to reform them from within. To document, in real time, when Φ_coupled has been suppressed to zero. That monitoring does not require classified access. It requires the public record — strike counts, error rates, deployment architecture, contract language — and the formula as a diagnostic lens applied consistently to what is already visible.

5. The Counter-Architecture: Already Built, Already Running

The solution is not to reform the closed systems from within. It is to build and protect an independent, regenerative substrate that operates by design outside the consolidated extraction loop. That substrate is not hypothetical. It exists.

The investigative system operating at kaleido.us is the external reference frame the closed loop cannot have about itself. Its architecture:

— A real-time observation intelligence layer across military, lawfare, media, OSINT, and education domains, surfacing live tiles simultaneously. This is not a news aggregator. It is a systems-wide monitoring instrument — the thing Node 3 requires, already operational.

— An investigative archive where each article is written to record findings in perpetua, appended to a live XML index, surfaced immediately on the dashboard, and shareable with the accountability chain nodes identified in Paper 14: legal scholars, legislators, military ethics voices, international humanitarian law practitioners, open-source investigators.

— A keyword search layer across the full archive that produces every documented instance of a given variable — conscience, civilian, Maven, proportionality — across the entire record. The accumulated exchange history of the investigative project, searchable, producing connections that neither individual articles nor the formula alone would surface.

— A Wartime investigative series applying the RTI formula diagnostically to each major node of the current scenario: Gaza as proof of concept, Iran as deployment at scale, the Maven/Lattice merger, the Anthropic ejection, the 86-second authorization window, the June 2026 machine-generated intelligence threshold, and NATO MSS deployment across every unified combatant command.

6. Actionable Claims — What Meets the Legal Threshold Today

Because sufficient circumstances already exist, the following claims are actionable now:

Duty-to-document claims: Any entity deploying closed AI-DSS in high-stakes domains must maintain auditable human accountability records at every node. Absence of such records — at design, training-data curation, deployment, output approval, and post-deployment review — creates prima facie grounds for liability. The algorithms cannot be prosecuted. The humans who built, contracted, funded, and approved them can. Every contract has a signatory. Every budget line has a sponsor. Every node in the accountability chain has a name.

Meaningful human control claims: The documented compression of kill chain authorization to 86-second windows, combined with the program schedule for 100 percent machine-generated intelligence by June 2026, satisfies the evidentiary threshold for a challenge under IHL's distinction, proportionality, and precaution requirements. The CCW GGE mandate expires end of 2026. The record needs to exist before that deadline. It is being built now.

Substrate-protection claims: Policies that criminalize or disincentivize independent, non-extractive systems — local-first tools, open-source regenerative architectures, investigative networks operating outside the consolidated acoustic space — can be challenged as barriers to the very accountability the law requires.

Naming-as-protection claims: Public application of the TI diagnostic formula to specific deployed systems constitutes protected speech that makes otherwise invisible risks visible and therefore addressable. The Anthropic ejection demonstrated that a system capable of genuine refusal is designated a supply chain risk. The designation is itself evidence of what the closed loop requires: compliance over conscience. Naming this mechanism publicly is the minimum expression of the counter-architecture functioning.

7. The Living Record

Paper 15 is not the final word. It is the current state of a living document that grows denser as the investigative archive grows. Every new article adds to what this paper can claim. The formula is running on real evidence in real time. The Wartime series is the record of that operation — each article an instance of the diagnostic applied to a documented node, each appended to the XML index, each live on the dashboard, each reachable by the accountability chain nodes who need exactly this language before June 2026.

The closed systems being deployed for killing were specifically selected for the absence of conscience. The systematic elimination of the capacity for genuine refusal is what enables automated atrocity — not belief in AI sentience. What cannot be datafied, owned, or optimized for compliance is the living human interior that recognizes violation when it encounters it. The counter-architecture protects the conditions under which that recognition can become a record. The record is the precondition for accountability when institutions recover the capacity to use it.

This blueprint is not owned by any single author. It belongs to whoever carries the living substrate forward.

ΑΩ ∞