Who is liable when the damage is caused by a decision made in milliseconds by a system whose causal chain runs through the designer, the operator, the certifier, the model trainer, and the regulatory framework that authorized it?

On January 24, 1978, the Soviet satellite Cosmos 954 scattered radioactive debris across 124,000 square kilometers of Canadian territory. It was pure hardware: no algorithms, no decisional autonomy, no causal complexity. The responsible party was identifiable; the applicable treaty existed. Canada still received half of what it claimed, with no acknowledgment of liability, three years later. If that case took three years to produce a partial settlement, what happens when the damage stems from an algorithm no one explicitly programmed to do what it did?

Chapter 9 of CLA builds the liability architecture of the Algorithmic Common Law: five layers differentiated by sphere of control — the system, the operator, the designer, the certifier, the regulatory framework — each with its own regime, identified subject, and precise activation mechanism. The weighting is asymmetric: violating a Threshold of Inviolability eliminates all liability caps, because some lines cannot have a price without ceasing to be lines. The chapter integrates the VEC fault taxonomy, the Sovereignty of Evidence, and the Algorithmic Dignity thresholds into a tiered fund system that guarantees compensation to victims before fault is determined — shifting the burden that ordinary law places on those least able to bear it.

The regime is not designed to compensate damage. It is designed to prevent it. Every presumption, every forensic audit, every risk-weighted premium exists so that operators build systems that never trigger the funds. Liability, distilled into six canonical axioms, will have served its purpose not when its mechanisms are activated, but when the architecture of consequences becomes an architecture of care.

🔹 CLA — Algorithmic Law for the Cosmos

Jesús Bernal Allende | Escuela del Deber-Optimizar y la Soberanía de la Evidencia

🌐 https://edo-os.com 🔗 https://www.linkedin.c

On September 2, 2019, the European Space Agency fired Aeolus's thrusters at 320 kilometers altitude to avoid Starlink 44. The collision probability had climbed to 1 in 1,000 — ten times ESA's action threshold. SpaceX, notified days earlier, did not maneuver. A bug in its internal alert system prevented operators from seeing the risk updates. ESA acted alone. No one violated any rule because there is no rule to violate: coordination between operators is negotiated by email, with no binding protocol, no traffic authority, no auditable record.

The incident was minor. The question it reveals is not. What regulatory regime applies to a navigation satellite making autonomous evasion decisions? The same as a life support system rationing oxygen in a Martian colony? The same as an algorithm adjudicating disputes between asteroid belt mining operators? The obvious answer is no. Current space law lacks the tools to articulate that difference.

Chapter 8 of CLA builds the functional taxonomy that space law needs and does not yet have. Four dimensions: function (what the system does), criticality (what happens when it fails), autonomy (how much human supervision it requires), and domain (where it operates). Five functional classes — from life support systems (Class A, existential criticality) to adjudication and governance systems (Class E). Four autonomy levels calibrated by physics: the 22-24 minute round-trip latency to Mars makes continuous ground control impractical, and no institutional design can change that constraint. The Regulatory Cube — the intersection of criticality × autonomy — determines applicable minimum requirements: the evidentiary level demanded, the intensity of VEC conditions, the configuration of dignity thresholds.

Three structural patterns emerge. First, the maximum-intensity diagonal: a Martian life support system with adaptive autonomy simultaneously mobilizes the full ANCLA triad at maximum intensity — not accumulated bureaucracy, but institutional response proportional to the highest conceivable risk. Second, the prohibition on existential-criticality with single-human supervision: when a system whose failure kills within minutes depends on one human, its safety is only as strong as that human's attention at the worst possible moment. Third, a deliberate asymmetry: the taxonomy does not penalize autonomy itself, but autonomy without supervision proportional to risk.

The taxonomy reveals that the core regulatory problem is not AI in the abstract but AI in context. Governing orbital traffic by email is not neutral omission — it is a political decision with identifiable beneficiaries. Without taxonomy, the operator who redesigns a life support system as a "data management platform" avoids the most demanding controls. Without taxonomy, the victim has no legal language to articulate the difference. Classification without consequences is nomenclature. Nomenclature without accountability is a catalog.

Chapter 9 will translate this taxonomy into accountability chains: who answers for what when a system of a given class fails.

🔹 CLA — Algorithmic Law for the Cosmos Jesús Bernal Allende | Escuela del Deber-Optimizar y la Soberanía de la Evidencia 🌐 https://edo-os.com 🔗 https://www.linkedin.com/in/jesus-bernal-allende-030b2795

Can a system be demonstrably efficient and radically unjust at the same time — without breaking a single rule it designed for itself?

In 2018, a hiring algorithm deployed by a major tech firm systematically screened out female candidates. There was no technical malfunction: the system optimized exactly what it was told to optimize. The results were auditable. The problem was that no one had encoded the constraint that human beings cannot be treated as variables to be eliminated from an optimization function.

Chapters 5 and 6 of CLA established the validity conditions and evidentiary standards of the Common Law Algorítmico. Both assumed the existence of constitutive constraints — limits that no optimization may cross. Neither formalized them. This chapter builds those constraints.

Algorithmic Dignity is not a philosophical extension of classical human dignity: it is its operational translation into the only language algorithmic agents understand — hard constraints that define the solution space before any calculation begins. Classical dignity operates ex post: a tribunal determines whether an act violated it. Algorithmic Dignity operates ex ante: the violation does not exist as a computable option.

The chapter formalizes seven Thresholds of Inviolability — organized across two levels (Alpha: absolute; Beta: subject to mandatory human escalation) — ranging from the prohibition on causing intentional death through optimization (U1) to the right to contest any algorithmic decision affecting fundamental rights (U7). For each threshold: converging philosophical foundations (Kant, UDHR, Jonas, Nussbaum), technical implementation specifications, and the precise consequences of violation. The relationship between VEC (Ch. 5) and Dignity is formalized as a lexicographic utility function: the thresholds do not participate in any cost-benefit calculation. No number of lives saved converts an intentional killing into an admissible option. The system does not choose not to do it; it simply cannot compute it as a choice.

As the chapter itself states: "A system that can prove it is efficient but cannot guarantee it is human has proved nothing."

🔹 CLA — Algorithmic Law for the Cosmos Jesús Bernal Allende | Escuela del Deber-Optimizar y la Soberanía de la Evidencia 🌐 https://edo-os.com 🔗 https://www.linkedin.com/in/jesus-bernal-allende-030b2795