Purpose and classification
Purpose, system, role, risk category and permitted use are defined case by case. No tool is generally authorised for every mandate.
Forskning og AI-styring
Technology as support. Human accountability.
AI, OSINT and data-analysis tools can accelerate the work; source validation, judgement and accountability remain human.
14 / Research and AI
Utgangspunktet
Each use of AI is qualified by purpose, data and risk.
Sentinel uses artificial-intelligence tools to support research, synthesis, scenario comparison and operational production. No automated output is treated as a fact, evidence or decision without human review proportionate to the mandate.
The governance model distinguishes Sentinel’s actual role, the system used and the use case. Not every use of AI is high-risk; prohibited uses or uses incompatible with the mandate are excluded, while uses subject to specific obligations are identified before activation.
01AI literacy
02Source traceability
03Human review
04Mandate disclosure
Operational controls
Human-in-command throughout the entire work cycle
Sources and counter-verification
Source provenance, date, authority and limitations are recorded; facts, inferences and scenarios remain separate and material conclusions are subjected to counter-verification.
Data and confidentiality
Personal data, confidential documents and trade secrets are not entered into external AI services without necessity, a lawful basis, appropriate contractual terms, security assessment and control of transfers.
Competence and supervision
People involved receive instructions and training appropriate to the system and risk. Human review must be substantive, documentable and capable of rejecting the output.
Transparency
Proportionate disclosure, not generic formulae
Professional mandates
Where AI is materially used as support in a professional activity, the client is informed clearly and understandably; human intellectual work remains predominant.
Public content
Synthetic or manipulated content is identified where required by law, including relevant cases involving deepfakes and texts on matters of public interest.
Decisions concerning individuals
Sentinel does not entrust decisions producing legal or similarly significant effects on individuals to exclusively automated processes. Any different case requires a dedicated legal and privacy assessment.
Declared limitations
Source version, coverage, assumptions, update date and material uncertainty accompany the output where they affect decision use.
Lawful research
Legitimate access, intellectual property and data discipline
OSINT does not mean unrestricted access
Open sources are used in compliance with terms of use, rights, access controls and the purpose of the mandate; circumvention of protections, pretexting and unauthorised access are excluded.
Text and data mining
Automated extraction and analysis respect lawful access, rights-holder reservations, copyright, database rights and statutory limits.
Trade secrets
Acquisition, use and disclosure must be lawful and consistent with NDAs, duties of confidentiality and contractual restrictions. Technical availability of a document does not make its use unrestricted.
Public data and research
Reuse of public-sector data follows licences, reuse conditions, data protection, third-party rights and source limitations; public availability of the data does not remove those constraints.
Institutions and authorities
Binding rules, guidelines and best practices remain distinct
European Union and Italy
Principal references: Regulation (EU) 2024/1689, GDPR, Italian Law 132/2025, and Italian rules on data, copyright, trade secrets and contractual liability.
AgID and public administrations
In public-sector mandates, applicable law, tender documents, the contract and finally adopted and relevant AgID guidelines apply. Consultation drafts are not treated as already-binding obligations.
AGCOM
AGCOM measures and obligations are relevant only where the activity, content or channel actually falls within the Authority’s regulated perimeter for media, communications, platforms or other sectors.
ANCI and the Ministry of the Interior
ANCI materials are best practices unless incorporated into acts or contracts. For the Ministry of the Interior, rules within its competence apply, including the TULPS licensing boundary: technical publications are not equivalent to general directives for private advisory work.
Executive Diagnostic
Automation does not replace accountability.
Each mandate defines permitted tools, usable data, review levels and disclosure obligations.
Start samtalen