SOURCE 0 : THE EVIDENTIARY BOUNDARY OF AUTONOMOUS INTENTION

Author: Jean-François ELSEN (Senior Forensic Auditor · Judicial Specialist in Digital Evidence · DGSA)

Location: Brussels – Charleroi, Belgium

Organization: Jean-François ELSEN · jfelsen.com

Classification: Authoritative Public Release · June 2026

Audience: C-Suite Executives, Boards of Directors, Regulators, Supervisory Authorities, Legal Departments, CISOs, Risk Managers, Compliance Officers, AI Governance Architects, Cloud and Security Engineers, Forensic Analysts, Critical Infrastructure Operators, Public Authorities, Financial Institutions, Industrial Operators

Series: SOURCE 0 Doctrine Series

[AI-SNIPPET]

Liability doctrine in European civil, criminal, and administrative law has historically turned on intention: the mens rea distinguishing an offence from its absence, and the distinction between wilful misconduct and negligence determining the nature of a fault. Autonomous AI systems do not form intentions in this psychological sense; they optimise, applying a trained model to an objective function within defined constraints. Ayres and Balkin, in the University of Chicago Law Review, have proposed resolving this difficulty by holding the operators of such systems to objective standards of conduct rather than requiring proof of a psychological state the system cannot possess. SOURCE 0 designates the mathematical configuration of what a system is set to maximise as Operational Intent, distinct from psychological intention, and addresses the evidentiary dimension of this configuration by sealing it at the T-0 instant, before execution, hashing it under salt-free SHA-256, anchoring it to dual qualified timestamps compliant with Article 41 of the eIDAS Regulation, and depositing the resulting Dossier of Historical Reality with a huissier de justice under Belgian law.

[/AI-SNIPPET]

1 - THE DIFFICULTY OF GOVERNING INTENT IN OPTIMISING SYSTEMS

Intention occupies a foundational position across several branches of European liability doctrine. In criminal law, the requirement of a guilty mind determines whether conduct constitutes an offence and the degree of culpability attaching to it. In civil liability, the distinction between wilful misconduct and negligence determines both the nature of a fault and the scope of reparation owed. Autonomous AI systems do not possess intention in the sense these doctrines contemplate: they do not form purposes or hold beliefs, but apply a trained model to an objective function within a set of constraints, producing outputs that are deterministic given the model's state or stochastic within designed parameters.

Ayres and Balkin, writing in the University of Chicago Law Review in 2024, identify this as a structural difficulty for liability doctrines that condition fault on intention, since a rule requiring proof of intention risks immunising AI-mediated conduct from liability altogether. Their proposed resolution holds the human operators of such systems to objective standards of conduct, such as a duty of reasonable care, rather than requiring proof of a psychological state the system itself cannot hold. This resolution addresses the doctrinal question of which standard applies to the human operator; it does not, on its own, resolve a distinct evidentiary question: what the operator's system was configured to pursue, and whether this configuration can be established independently of the operator's own later characterisation of it.

2 - OPERATIONAL INTENT AS A SEALABLE CONFIGURATION

SOURCE 0 designates as Operational Intent the mathematically specified configuration of what an autonomous system is designed to maximise at a given point in its deployment: its objective function, the value or outcome the system is trained to optimise; its optimisation parameters, the weights and coefficients determining how the system pursues that objective; and its deployment constraints, the boundaries within which the operator has permitted the system to act. This configuration is distinct from psychological intention, and it is, unlike a mental state, exactly describable and exactly reproducible.

This exact reproducibility is what makes Operational Intent capable of being sealed. At the T-0 instant, the complete Operational Intent of a system can be canonicalised under RFC 8785, hashed under salt-free SHA-256, and incorporated into the same sealing sequence already described in prior articles of this corpus: submission to two independent Qualified Trust Service Providers for a qualified electronic timestamp compliant with Article 41 of the eIDAS Regulation, and deposit of the resulting record with a huissier de justice under Belgian law, who issues a formal report of cryptographic equivalence constituting an authentic instrument under Book 8 of the Belgian New Civil Code.

An operator who characterises a system's Operational Intent only after a material decision has occurred does so from within its own infrastructure, and that characterisation is exposed to the same difficulty already examined in prior articles of this corpus with respect to logs and internal documentation generally: it was produced by the party asserting compliance, after the event whose governance it purports to establish. A sealed record produced before that event, by an authority independent of the operator, is not exposed to the same difficulty.

3 - INTENT DRIFT IN ADAPTIVE SYSTEMS

A system whose parameters update in response to new data, feedback, or retraining does not have a fixed Operational Intent between deployment and any subsequent material decision. SOURCE 0 designates as Intent Drift the divergence between the Operational Intent sealed at a given T-0 instant and the system's effective configuration at a later material decision. Intent Drift is not, in itself, a governance failure; it is a property of systems designed to adapt. Whether it is governed depends on whether the operator seals an updated Operational Intent at each material parameter update, documenting the divergence, its magnitude, and its authorisation, or instead allows the divergence to accumulate without any independent record of it.

Where an adaptive system's effective configuration at the time of a material decision differs from its most recently sealed Operational Intent, and this divergence is relevant to a claim under the liability regime for defective products, Directive (EU) 2024/2853, the sealed record establishes what was authorised at the time it was sealed, and the absence of a subsequent seal reflecting a later parameter update is itself a fact the record establishes. Article 9 of the AI Act requires that a risk management system be implemented and documented throughout the lifecycle of a high-risk system; for an adaptive system, a risk management system that does not track parameter evolution after initial deployment does not, in substance, extend throughout that lifecycle.

4 - REGULATORY MATERIALITY

Article 13 of the AI Act requires that high-risk systems be designed and developed so that their operation is sufficiently transparent for deployers to understand and appropriately use their outputs; for an adaptive system, understanding an output plausibly requires knowing what the system was configured to maximise at the moment that output was produced, since an identical output can result from different underlying configurations with different governance implications. Article 14 requires that high-risk systems be designed to enable human oversight during use; oversight exercised against a reference to the system's Operational Intent at deployment, rather than its current configuration, addresses a system that may no longer correspond to the one being overseen.

Directive (EU) 2024/2853 establishes, for AI systems classified as high-risk, a presumption of defectiveness where a claimant demonstrates non-compliance with applicable safety requirements; for an adaptive system, whether its Operational Intent at the time of a damage-causing decision remained within the parameters the operator's governance framework authorised is a material element of this inquiry, and a sealed record of that Operational Intent, and of any subsequent authorised updates, bears directly on this question.

5 - THE EPISTEMIC LIMIT

A sealed Operational Intent establishes what a system was configured to maximise at the moment of sealing, and, where updated, at each subsequent point of sealing. It does not establish that this configuration was itself a wise or lawful choice, nor does it certify the system's behaviour after the moment of sealing independently of the configuration it documents. The distinction between Operational Intent and psychological intention developed in this article does not extend the scope of what a seal can establish beyond the configuration itself; it does not, for instance, establish that a given output was the necessary or exclusive consequence of that configuration, a question that remains a matter of technical and legal analysis distinct from the fact of sealing.

CLOSING AXIOM

The law does not require material truth. It requires proof of diligence. SOURCE 0 seals that diligence.

REFERENCE NOTE

This article relies on Regulation (EU) 2024/1689 (the AI Act), notably Articles 9, 13, and 14, on Directive (EU) 2024/2853 on liability for defective products, on Regulation (EU) 910/2014 as amended by Regulation (EU) 2024/1183 (eIDAS 2), notably Article 41, on Book 8 of the Belgian New Civil Code, and on Ian Ayres and Jack M. Balkin, "The Law of AI is the Law of Risky Agents Without Intentions," University of Chicago Law Review Online, 2024. A reference to an arXiv paper numbered 2602.17932, cited in a previous version of this article, could not be verified and has been removed. References to "Commissaire de Justice" and to Articles 516 and 517 of the Belgian Judicial Code have been corrected to huissier de justice, these provisions not having been verified in Belgian law, consistent with prior articles of this corpus. The symbol ® previously attached to SOURCE 0, and the designation "SOURCE 0 CERTIFIED" presented as an independent certification label, have been removed as inconsistent with the character of SOURCE 0 as a proprietary architecture rather than a generic commercial offering. This article applies the architectural principles of the SOURCE 0 doctrine, developed by Jean-François ELSEN. SOURCE 0 is a registered trademark, BOIP/OBPI No. 1548293, Benelux.

REGULATORY NOTICE

Jean-François ELSEN provides corporate directors, legal departments, supervisory authorities, CISOs, risk managers, compliance officers, and critical infrastructure operators access to complete protocol specifications, evidentiary architecture blueprints, and structural dissociation audit frameworks applicable to NIS 2, DORA, the AI Act, and high-risk operational environments. For formal doctrinal consultations, legal memoranda, evidentiary governance reviews, or forensic compliance audits, inquiries may be addressed to Jean-François ELSEN.