AI transparency,
without theatre.
What AIA SecTest does, what it cannot conclude, how humans remain responsible, and which EU AI Act evidence still requires internal or legal validation.
Runtime disclosure
Human-reviewed advisory tool
The deployed audit engine uses deterministic checks and structured remediation templates. It does not currently invoke a conversational or generative AI model, and it never makes the final security or legal decision.
Preliminary role
Software service provider
AIA Systems provides the scanner. Formal AI Act actor roles depend on whether later versions add an AI system or GPAI integration and require legal validation.
Preliminary classification
Not identified as high-risk
Current intended use is advisory security analysis. It is not designed for an Annex III decision and must not be used as an autonomous high-risk decision system.
Human control
Review before action
A person must verify findings, decide remediation, assess applicability, and approve any deployment or legal conclusion.
01
AI system inventory
The current public system record. Items marked preliminary must be confirmed by the accountable owner and qualified counsel.
System
AIA SecTest
Public owner
AIA Systems
Lifecycle
Production pilot · public web service
Intended users
Website owners, authorised testers, developers and security teams
Intended purpose
Non-destructive public-surface security checks and observable EU AI Act evidence review
Decision significance
Advisory only; no autonomous approval, denial, ranking or enforcement decision
Data handled
Target URL and bounded public HTTP/TLS responses; no credentials requested
Current model/provider
No GPAI or generative model is invoked by the deployed audit engine
Primary integration
Rate-limited outbound HTTP/TLS requests to the authorised target
Availability
Public web; users must confirm ownership or explicit permission
02 · Intended use
What the system is for
Identify externally observable security weaknesses using bounded, non-destructive checks.
Review publicly visible EU AI Act transparency and governance signals.
Create structured remediation prompts for a human-supervised coding workflow.
Support—not replace—penetration testing, source review, governance evidence and legal assessment.
03 · Prohibited use
What the system must not be used for
Testing a target without ownership or explicit authorisation.
Credential attacks, destructive access, denial of service or attempts to extract private data.
Claiming certification, guaranteed security or legal compliance from the score.
Autonomous hiring, education, credit, biometric, law-enforcement or other high-impact decisions.
04 · Capabilities and limitations
Read the score with the right amount of trust.
A high score means the executed checks found fewer visible problems. It does not prove that untested code, authenticated paths, business logic, databases or internal governance are safe.
Can observe
TLS, headers, public cookies, CORS, exposed files, client-visible token patterns and bounded public canaries.
Cannot observe
Source-level flaws, row-level security, private APIs, internal policies, staff competence or real deployment context without additional evidence.
Accuracy boundary
False positives and false negatives remain possible. Every finding and compliance observation requires human verification.
Change trigger
Adding an LLM, synthetic-media generation, automated decisions or high-risk use requires a new classification and updated notice before launch.
05 · Human oversight
Challenge, correct or escalate a result
Pause remediation when evidence is unclear or the proposed fix could weaken another control.
Verify evidence against code, infrastructure and authenticated workflows.
Document accepted risk, rejected findings and the responsible approver.
Escalate suspected incidents, incorrect reports or legal-impact questions to the accountable AIA Systems contact.
Accountable contact route
AIA Systems support and escalation
For inaccurate findings, AI Act evidence corrections, accessibility needs, security incidents or a human review request, use the official AIA Systems contact channel. Include the audited URL, scan time and finding ID; never send passwords or secrets.
Public owner: AIA Systems · Operating jurisdiction presented for validation: Greece/EU · Legal registration details require owner confirmation.
06
AI literacy programme
A role-based programme design aligned to Article 4. Publication of the design does not prove training completion; attendance and assessment records remain internal evidence.
Everyone
AI basics, opportunities, limitations, authorised use and incident reporting
Developers
Secure AI integration, prompt/data risks, model limitations, evaluation and logging
Security reviewers
Scanner interpretation, false positives/negatives, human verification and escalation
Product & leadership
Provider/deployer roles, risk classification, fundamental rights and launch gates
Support & operations
User explanations, complaint handling, evidence preservation and human review
Completion
Evidence required
Effectiveness check
Scenario assessment
Refresh cadence
Annual + material change
07
Compliance evidence register
Public readiness register. “Published” means the evidence is visible; it does not independently establish legal compliance.
| Workstream | Status | Owner | Closure evidence |
|---|---|---|---|
| Article 5 screening | Published | Product + legal | Prohibited-use policy and launch-blocking review |
| Article 4 AI literacy | Programme defined | Operations | Role matrix, completion records and refresh log |
| Article 50 transparency | Current scope documented | Product | Runtime behavior, disclosure trigger and future-change gate |
| Articles 13–14 information and oversight | Published | Product + security | Limitations, human review and challenge route |
| Article 6 / Annex III classification | Legal validation required | Legal + product | Intended purpose, deployment context and signed classification |
| Article 53 GPAI assessment | Not applicable to current architecture | Engineering + legal | Architecture record; reassess before adding model training or provision |
08 · GPAI assessment
Not a GPAI model provider in the current architecture
AIA SecTest does not train, place on the market, or provide a general-purpose AI model. Article 53 provider documentation is therefore marked not applicable to the current deployed architecture—not permanently exempt. Adding a trained model, fine-tuned model offered downstream, or model API under AIA Systems' control triggers reassessment.
09 · Content provenance
No synthetic-media generation today
The service produces structured text prompts from deterministic finding templates. It does not generate image, audio, video, deepfake or public-interest content with a generative model. Any future generative feature must add visible disclosure, machine-readable marking where applicable, export preservation tests and a revised Article 50 assessment before release.
Open validation items
• Confirm legal entity name, registration details, jurisdiction and accountable contact.
• Obtain signed Article 6 and Annex III classification from product and counsel.
• Record AI-literacy completion and effectiveness evidence by role.
• Approve incident, complaint, appeal and document-retention procedures.
Document control
Owner and change history
Document owner: AIA Systems product and security lead · Review cadence: at least annually and before any material AI, purpose, data or risk change.
Authoritative references
This notice supports transparency and evidence collection. It is not legal advice, a conformity assessment, certification, or a substitute for qualified review of the actual organisation, system and deployment context.