AI may assist. People remain accountable.

Scope

This statement describes AI-assisted work used to develop and maintain the public 4SI website, research service, code and visual materials. It distinguishes internal assistance from the behavior of the live public service.

The current public website, Presence Risk Index and Exposure Radar do not call a generative-AI service when a visitor reads, searches or filters a page. Visitor browsing and filter activity are not sent to an AI model, and public conclusions are not generated on demand.

Where AI may assist

Approved AI-assisted tools may support source triage, field normalization, duplicate review, translation support, comparison, drafting, code generation, testing, accessibility review, design exploration and visual production.

Assistance is task-specific. It does not give a model authority to approve a source, establish a fact, assign Critical status, make a legal conclusion, publish an allegation or determine an outcome about a person or organization.

Evidence decides

• AI output is not a source, fact or independent confirmation.
• Published factual claims must map to cited evidence.
• Missing facts are not filled through generation.
• Source facts, observed signals, validated patterns, 4SI assessments, modeled exposure and uncertainty remain distinct.
• Material calculations, links and source relationships are checked against the underlying record.
• Critical Exposure status requires documented human approval.
• Corrections follow the published Corrections Policy.

Human review does not make model output infallible. It creates an accountable decision point and a correction path.

Generated and modified media

4SI may use generative tools for product visualization, architectural visualization, concept imagery, diagrams or controlled image modification. A visual that could reasonably be mistaken for documentary evidence of a real event, person, place or deployed facility is labeled in context as an AI-generated image, visualization, concept render, composite or equivalent.

Generated or materially modified visuals are not presented as incident evidence. Where technically supported and appropriate, 4SI preserves or adds machine-readable provenance or content credentials. Captions and adjacent context remain the primary human-readable disclosure because metadata may be removed by delivery systems or downstream copying.

For AI-generated or manipulated text on a matter of public interest, 4SI applies human editorial review and responsibility and provides an additional disclosure where required by law or where omission could materially mislead the audience.

Data and provider boundary

4SI maintains an internal record of approved AI providers, model or tool family, material task and review owner because providers and model versions can change. Public visitor analytics, contact-form contents, passwords, authentication secrets, unreleased vulnerability details and unnecessary personal data are not approved inputs to general-purpose AI tools.

Any use involving confidential, personal or security-sensitive information requires an authorized environment, a defined purpose, minimization, access control, retention review and any necessary contractual protection. AI providers do not receive public-site analytics merely because a visitor browses the website.

What the public service does not do

• no biometric identification of Index or Radar visitors;
• no individual automated decision producing legal or similarly significant effects;
• no claim that AI output is objective, neutral, complete or error-free;
• no automated publication of unverified allegations or Critical Exposures;
• no live generative scoring during an ordinary page visit.

Governance and changing rules

4SI reviews this statement when a material workflow, provider, public feature or legal requirement changes. The owner of the relevant work remains responsible for validation, publication and correction. Material incidents or failures are escalated according to their impact.

The EU AI Act’s Article 50 transparency obligations apply from 2 August 2026. 4SI uses the final European Commission transparency guidance, the official Regulation and applicable codes or standards as reference points while assessing each use case. This statement describes current controls; it does not make a blanket claim of compliance with every AI law in every jurisdiction.