Specifying the expected behaviour
What the agent must do, what it must never do, when it must stop and ask. Writing the system instruction files and the scope rules.
Compliance and reliability of agentic AI systems
An agent should reason like a competent professional: refuse when unsure, escalate instead of inventing, ask for a clarification rather than assume, stay within its scope. At the same time it must be clearly identified as a machine, in line with the transparency obligations of the European regulation on artificial intelligence.
These two requirements do not conflict. The first concerns the quality of judgement, the second the identity of the sender.
In brief
AGGIL specifies, evaluates and documents the behaviour of AI agents: refusal and escalation rules, evaluation suites, compliance points placed in the architecture. AGGIL applies to AI agents the validation methods for systems whose behaviour cannot be enumerated, practised since 2013 on model-based testing and autonomous driving. AGGIL does not certify the compliance of a system.
Validation methods
AGGIL applies to AI agents the validation methods for systems whose behaviour cannot be enumerated, practised since 2013 on model-based testing and autonomous driving.
Talks and publications on autonomous vehicle validation (2014-2021)
What we do
Refusal and escalation rules
The cases where the agent hands control back instead of producing an answer, and the person or step it hands it back to.
Behaviour evaluation set
Test cases written before going live, the expected behaviour for each, measurement of deviations at every version, behavioural non-regression tests.
Compliance points placed in the architecture
Where the notice to the person, the marking of generated content, human validation, the logs and their retention period are placed. These points are decided at design time.
Documentation and traceability
Instruction files versioned in the repository, session logs kept, dated architecture decisions.
Formats
- Scoping — 5 days.
- Prototype — 15 to 25 days.
- Production rollout, on quotation.
Delivered : system instruction files, scope rules, behaviour evaluation set, architecture diagram showing the human validation points and the logs, dated architecture decisions.
Governance assignments may use ISO/IEC 42001:2023 (artificial intelligence management system) as a framework: gap assessment against its requirements, documentation, audit preparation. AGGIL is not a certification body; certification is the remit of an accredited body.
The link with the research tax credit
Characterising an expected behaviour, writing an evaluation set, measuring deviations, correcting and starting again is an experimental approach. When the result is not known in advance and the solution cannot be deduced from the state of knowledge, this work may fall under research and development; substantiating it requires the instruction files, the evaluation sets and the successive versions to be dated and kept.
Research and innovation tax credit: software and AI projectsSélectionner, assembler, adapter : ce que le développement par agents d’IA laisse au crédit d’impôt recherche. Le Village de la Justice, 21 September 2026.
Transparency obligations
Article 50 of Regulation (EU) 2024/1689 covers transparency obligations, in particular informing people who interact with an artificial intelligence system and marking artificially generated content. These obligations have applied since 2 August 2026. For systems generating synthetic content placed on the market before 2 August 2026, the marking obligation (Article 50(2)) applies by 2 December 2026 at the latest (Article 111(4), inserted by Regulation (EU) 2026/1744).
The obligations for high-risk AI systems listed in Annex III of Regulation (EU) 2024/1689 apply from 2 December 2027 (Article 113, as amended by Regulation (EU) 2026/1744 of 8 July 2026).
Our role is to design the architecture so that these obligations can be met, and to document the choices made. We do not certify the compliance of a system.
An agent to specify, a behaviour to correct
Describe the context and the question. The answer will state what is feasible, in what form, and in how many days.
Get in touch