Investigating Serious AI Incidents: Evidentiary Standards, Reporting and Accountability

Investigating Serious AI Incidents: Evidentiary Standards, Reporting and Accountability

Published
November 19, 2026
Author
Khrabrykh S. A.
Original language
English

The report examines the investigation of serious incidents involving artificial intelligence systems and the evidentiary sufficiency of the explanations given for them. Its central question is which observations confirm that a specific component contributed to an adverse outcome, such as a wrongful identification or an erroneous restriction of access, which alternative explanations remain open, and how the effect of a correction can be verified. The methodology separates the reconstruction of the event, the testing of the causal mechanism and the legal characterisation: neither an algorithmic assessment nor a management-system certificate amounts in itself to a finding of fault.

The study draws on digital forensics, causal analysis, system safety and statistics. It covers the preservation of digital evidence and chain of custody, timeline reconstruction when clocks do not match, model logs, prompts and retrieval context, whether human oversight is real or merely nominal, reproduction and counterfactual testing, signs of targeted attacks such as prompt injection and memory manipulation, and the distinction between model failure and process failure.

The legal analysis, current as of 22 October 2026, treats separately serious incidents under Articles 55 and 73 of the AI Act, personal data breaches under the GDPR, significant incidents under NIS2 and major ICT-related incidents under DORA, establishing for each regime the legal threshold, the addressee and the point from which the notification period runs. The report also considers administrative, criminal and civil proceedings, the rights of affected persons and typical risk areas: sanctions screening, international wanted-person searches and extradition, AML/KYC, credit scoring, medical AI and deepfakes presented as purported evidence.

The main conclusion is that an investigation must reconcile three distinct results: a verifiable account of events, a substantiated causal mechanism and a legal assessment of the consequences. None of these methods on its own establishes the fault of a particular person. Closure of an incident must be confirmed on its technical, procedural and remedial sides: fixing the model does not end the work while an erroneous record continues to circulate or a restriction of rights remains in place.

For ARGA, the report proposes a causal investigation protocol with a register of research questions, a map of available and lost data, a matrix of alternative hypotheses, reproduction tests and completion criteria, together with five quality standards for independent assessment. The fifty annexes are expert forms and training scenarios, not regulator-approved templates or real cases. The report contains no experiment on an actual case, does not establish the effectiveness of the methodology, and does not replace the assessment of a lawyer, forensic expert or supervisory authority.

Full text of the document
ARGA Observatory

Citation Rules

  1. Mandatory source attributionWhen using ARGA Observatory materials, the full name must be cited.
  2. Date and version indicationFor analytical reports, the year of publication is mandatory.
  3. Link to the originalElectronic materials must be accompanied by an active link to the official website.
  4. Context preservationCitations must not be shortened or altered in a way that distorts the original meaning.
  5. Note on adaptationIf the text is abridged or translated, state: "adapted from ARGA Observatory report".
  6. No commercial use without written permission from the organization
  7. Data accuracy preservationCharts and tables must be reproduced without changes.
Confidential enquiry

Ask the ARGA experts

Four short steps. We reply within one business day.

Step 1 of 4
What situation are you facing?
What stage is the case at?
Describe the situation
You can skip this field.
How should we reach you?
Your enquiry goes straight to ARGA experts. We do not publish it or pass it to third parties.
Official registration
ARGA in international registries
UNGM ID
1232417
ARGA is registered in the United Nations procurement system.
View registration
UN Global Compact
213173
Participant of the UN Global Compact.
ADB GMS Number
049963
Asian Development Bank — registered participant.
Scroll to Top