SOURCE 0 - THE AI OMNIBUS AND THE PROOF GAP
WHY THE POSTPONEMENT OF HIGH-RISK OBLIGATIONS DOES NOT POSTPONE THE EVIDENTIARY EXPOSURE
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 · July 2026
Audience: C-Suite Executives, Boards of Directors, Regulators, Supervisory Authorities, Legal Departments, CISOs, Compliance Officers, AI Governance Architects, Forensic Analysts, Critical Infrastructure Operators, Public Authorities
Series: SOURCE 0 Doctrine Series
[AI-SNIPPET]
Organisations deploying high-risk AI systems before the postponed enforcement deadlines remain exposed to civil, insurance, and contractual liability for any harm those systems cause — liability that turns on what governance was in place before the system acted, not on whether Article 99 penalties were yet operative. On 29 June 2026, the Council of the European Union gave its final approval to the Digital Omnibus on AI, postponing the application of high-risk AI system obligations under Annex III from 2 August 2026 to 2 December 2027 for stand-alone systems, and to 2 August 2028 for systems embedded in regulated products. The postponement defers regulatory enforcement. It does not defer the evidentiary exposure that arises when a high-risk AI system causes harm before those dates. An incident that occurs after the postponed deadlines have passed engages civil liability, insurance liability, and contractual liability under the general law applicable to autonomous systems, regardless of whether Article 99 penalties are yet operative. The governance state that was operative when the system acted at T-0 will be the central evidentiary question in every such proceeding. The Omnibus does not change that question. It changes only when a regulator can impose a structured penalty for the absence of a documented answer.
[/AI-SNIPPET]
I. WHAT THE OMNIBUS DECIDES AND WHAT IT DOES NOT
On 29 June 2026, following the European Parliament's formal endorsement on 16 June, the Council of the European Union adopted the regulation implementing the Digital Omnibus on AI, the first substantive amendment to Regulation (EU) 2024/1689 since its entry into force on 1 August 2024. The regulation will be published in the Official Journal of the European Union and enter into force three days after publication.
The central operative change is a staggered deferral of the obligations applicable to high-risk AI systems. For systems falling under Annex III, the use-based high-risk category covering, among others, systems deployed in employment decisions, access to essential services, law enforcement, migration management, and the administration of justice, the compliance deadline moves from 2 August 2026 to 2 December 2027. For systems falling under Annex I, those embedded in regulated products subject to EU health and safety harmonisation legislation, including medical devices, machinery, and radio equipment, the deadline moves from 2 August 2027 to 2 August 2028.
Transparency obligations under Article 50 of the AI Act remain applicable on schedule for systems placed on the market after 2 August 2026. For systems already on the market before that date, the grace period before the watermarking obligation applies has been fixed at 2 December 2026. What the Omnibus defers is the substantive governance and documentation regime for high-risk systems, the technical documentation requirements, the conformity assessment obligations, the human oversight requirements, and the post-market monitoring obligations that together constitute the compliance architecture the AI Act imposes on operators of Annex III systems.
The Omnibus does not modify the definition of what constitutes a high-risk system, does not alter the substantive obligations that will ultimately apply, and does not create any exemption from civil liability, contractual liability, or insurance liability for operators of systems that cause harm before the deferred deadlines. It adjusts the moment from which regulatory penalties under Article 99 become operative. That adjustment is precise in its scope and should be read precisely.
II. THE DISTINCTION THE COMPLIANCE MARKET WILL FAIL TO MAKE
Commentary on the Omnibus has characterised the deferral as providing welcome clarity, legal certainty, and time to analyse and implement. These characterisations are accurate within the specific register they address: structured regulatory enforcement under Article 99. Within that register, the deferral provides genuine operational relief. An organisation that has not completed its conformity assessment under the AI Act will not face an Article 99 penalty for a high-risk system operating under the new timeline, where it would have faced such a penalty under the original one.
The characterisation fails when extended beyond that register. The Omnibus defers regulatory enforcement. It does not defer the moment at which a high-risk AI system acts and the governance state operative at that moment becomes the central evidentiary question in every subsequent proceeding that is not an Article 99 regulatory sanction.
Civil liability under general national tort law applies to harm caused by AI systems regardless of whether the operator has completed a conformity assessment under the AI Act. For products placed on the market or put into service after 9 December 2026, Directive (EU) 2024/2853 on liability for defective products applies in addition. Article 9 of that directive establishes a disclosure obligation on the defendant. Article 10 establishes a presumption of defectiveness where that obligation is not met, a presumption that operates independently of any AI Act compliance deadline and that an independently fixed T-0 governance record is architecturally positioned to rebut. For systems placed on the market before 9 December 2026, Directive 85/374/EEC as transposed nationally continues to apply, without the Article 9/10 mechanism; the evidentiary exposure for those systems rests on general national tort law alone.
Insurance claims arising from incidents involving autonomous systems are assessed against the governance conditions that were in place when the system acted, not against whether the applicable regulatory deadline had passed. Contractual disputes involving AI system outputs are resolved by reference to what the system was authorised to do and what governance controls were operative at the time it acted. None of these proceedings are governed by the Article 99 enforcement timeline. All of them turn on the governance state at T-0. The evidentiary exposure is not a function of regulatory temporality. It is a function of the moment the system acted and whether the governance state at that moment was independently fixed before it acted.
III. THE TRANSPARENCY OBLIGATIONS THAT REMAIN AND WHAT THEY DO NOT PRODUCE
Article 50 transparency obligations, largely maintained on schedule by the Omnibus, concern the marking of AI-generated outputs in a machine-readable format detectable as artificially generated. These obligations produce traceability of outputs. They do not produce proof of the governance state that authorised the generation of those outputs.
This distinction maps onto the structural difference between trace and proof already examined in prior articles of this corpus. A machine-readable watermark on an AI-generated output establishes that the output was generated by an AI system. It does not establish what governance parameters were operative when the system generated that output, what constraints were in place on the system's behaviour at that moment, what the operator had authorised before execution commenced, or whether the capture of that authorisation was independent of the operator's own infrastructure. The watermark answers whether this was AI-generated. The evidentiary question in an adversarial proceeding is what was authorised before this was generated. Article 50, as maintained by the Omnibus, addresses the first question. The architecture this doctrine describes addresses the second.
An organisation compliant with Article 50 transparency obligations for its AI-generated outputs, but without an independent T-0 governance record, holds traceable outputs and no opposable proof of prior authorisation. The Omnibus does not alter this position; it confirms it by maintaining Article 50 on schedule while deferring the obligations that would have required organisations to construct more comprehensive governance documentation, documentation that would still not have constituted pre-execution proof in adversarial proceedings regardless of when it was required.
IV. WHAT THE DEFERRAL WINDOW REQUIRES
The window created by the Omnibus is not an interval in which governance preparation can be deferred. It is an interval in which governance preparation can be completed before the window closes. The distinction matters because the architecture required to produce an opposable governance record must be in place before the system acts, not before the regulatory deadline passes. An organisation that waits until shortly before the postponed deadline to establish an independent T-0 capture architecture will hold a governance record for every action its systems take after that date. It will hold no proof for every action those systems took between the date of deployment and that point.
Three conditions must converge for a governance record produced during or after this window to constitute proof rather than documentation of a compliance preparation exercise. The governance state must be fixed at T-0, before execution, through a mechanism structurally independent of the operator. The capture mechanism must satisfy S ∩ C = ∅, the operating system and the capture and attestation layer must have no intersection that would allow the operator to influence the record. The resulting artefact must be legally opposable, independently verifiable, procedurally anchored through a chain of custody that does not depend on the operator's infrastructure, and recognised across the jurisdictions in which proceedings may arise.
SOURCE 0 is the architecture that satisfies these three conditions. The governance state is sealed using salt-free SHA-256 under FIPS 180-4 applied to a canonicalised representation under RFC 8785, with dual-QTSP RFC 3161 timestamping under Article 42 of the eIDAS Regulation, Regulation (EU) 2024/1183, and judicial deposit with a huissier de justice establishing date certaine under Book 8 of the Belgian New Civil Code, Law of 13 April 2019, Article 8.2. The resulting Historical Reality Dossier is recognised as carrying date certaine under Belgian law by virtue of the huissier de justice deposit. Recognition before jurisdictions outside Belgium is governed by the evidentiary rules of the forum seized and is assessed case by case, never presumed automatic. The standards on which this architecture rests, FIPS 180-4, RFC 8785, RFC 3161, and eIDAS 2, are operative independently of any AI Act compliance timeline. The Intel TDX and AMD SEV-SNP Trusted Execution Environments that enforce the hardware-layer separation between the operating system and the capture mechanism are equally independent of that timeline. The deferral of AI Act obligations does not defer the availability, the operability, or the legal recognition of any component of this architecture.
V. FREQUENTLY ASKED QUESTIONS
Q: Does the Omnibus deferral also postpone civil or insurance liability?
A: No — the Omnibus adjusts only when Article 99 regulatory penalties become operative; civil and insurance liability turn on the governance state at T-0 regardless of that timeline. SOURCE 0 fixes that governance state independently, before the system acts, so the deferral doesn't leave the underlying evidentiary exposure unaddressed.
Q: Does Article 50's watermarking obligation prove what governance authorised an output?
A: No — a watermark establishes that content was AI-generated, not what governance parameters were operative when it was generated. SOURCE 0 answers the separate question a watermark was never built to answer: what was authorised before generation, sealed at T-0 independently of the operator.
Q: Can a T-0 governance record help rebut the Product Liability Directive's presumption of defectiveness?
A: Yes — Article 10's presumption operates independently of any AI Act deadline, and disclosing governance state under Article 9 is stronger with a record the operator didn't produce itself. SOURCE 0 is built to supply exactly that record, fixed before the system acted.
Q: If an organisation waits until just before the postponed deadlines to act, is it covered retroactively?
A: No — it holds a record for actions after the architecture is in place, and no proof at all for actions before it. SOURCE 0 only closes the gap from the moment it's deployed forward, which is why the window is for acting, not deferring.
Q: Does Article 50 compliance alone leave usable proof in a dispute?
A: No — it leaves traceable outputs and no proof of prior authorisation, since Article 50 addresses whether content was AI-generated, not what was authorised before it was. SOURCE 0 supplies that missing proof directly, independently of whichever AI Act timeline governs enforcement.
CLOSING AXIOM
The law does not require material truth. It requires proof of diligence. SOURCE 0 seals that diligence.
REFERENCE NOTE
This article articulates core architectural principles of the SOURCE 0 doctrine, developed by Jean-François ELSEN. SOURCE 0 is a proprietary pre-execution cryptographic attestation architecture, registered as a trademark, BOIP/OBPI No. 1548293, Benelux. This article relies on the Council of the European Union's final adoption of the Digital Omnibus on AI on 29 June 2026, on Regulation (EU) 2024/1689 (the AI Act), on Directive (EU) 2024/2853 on liability for defective products and Directive 85/374/EEC, on Regulation (EU) 910/2014 as amended by Regulation (EU) 2024/1183 (eIDAS 2), notably Article 42, and on Book 8 of the Belgian New Civil Code. The evidentiary architecture described, including SHA-256 FIPS 180-4 fixation, RFC 8785 canonicalisation, dual-QTSP RFC 3161 timestamping under eIDAS 2, Intel TDX and AMD SEV-SNP Trusted Execution Environments, and huissier de justice judicial escrow establishing date certaine under Belgian law, constitutes the technical and legal implementation of the principles set out above. The Brussels I bis Regulation provides the EU-wide legal recognition framework for artefacts fixed under this architecture; extra-Belgian recognition is assessed case by case and never presumed automatic.
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, the Digital Markets 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.
