Free tool

EU AI Act Article 50 Scanner

Enter a public AI product URL for observable EU AI Act Article 50 signals, or submit authorized private workflow evidence when the relevant disclosure, export or provenance record sits behind login.

View full Scorecards EU AI Act Article 50 hub

EU AI Act Article 50 applies from 2 Aug 2026Technical evidence, not legal advice

EU AI Act Article 50 scanner

Implement, test and prove your EU AI Act Article 50 transparency controls.

Find likely EU AI Act Article 50 surfaces, map them to applicable controls, implement fixes, test them continuously, and generate evidence for counsel, buyers, and engineering leaders.

Scan my AI product

Implementation reportexample-ai-product-com

Counsel review

Surfaces1

Controls0/5

Review7

CI pass2/5

1 AI surfaces mapped. 0 controls implemented. 7 awaiting counsel.Evidence record ready to share after runtime execution.

Shareable reportisgenai.com/report/example-ai-product-com

Scope first

EU AI Act Article 50 applicability wizard

Customer facts feed the rule engine before any scan finding is interpreted.

Organisation roleDirectly interacts with peopleAI context is obviousManipulates existing content

Generated modalities

TextAudioImageVideo

Deepfake-like content possiblePublic-interest text publishedHuman editorial responsibilityAssistive editing onlyEmotion or biometric featureOn market before 2 Aug 2026Code of Practice path

Live scope matrix

No single green badge. Just the EU AI Act Article 50 state.

EU AI Act Article 50(1)

Direct human-AI interaction disclosure

CUSTOMER FACTS REQUIRED

The scan cannot determine direct interaction from public signals alone.

CUSTOMER FACTUAL CONFIRMATION REQUIRED

EU AI Act Article 50(2)

Machine-readable marking and detection

NOT INDICATED

The Code-of-Practice path is not confirmed, so the report keeps both Code and alternative-measures evidence visible.

TECHNICALLY UNVERIFIABLE

EU AI Act Article 50(3)

Emotion recognition and biometric categorisation notice

NOT INDICATED

No emotion recognition or biometric categorisation fact is currently indicated.

TECHNICALLY UNVERIFIABLE

EU AI Act Article 50(4)

Deepfake and public-interest content labelling

NOT INDICATED

Deepfake or public-interest publication facts are contextual and should be reviewed before runtime controls are disabled.

TECHNICALLY UNVERIFIABLE

EU AI Act Article 50(5)

Accessible and context-appropriate labels

CUSTOMER FACTS REQUIRED

Any human-facing disclosure or label should be clear, distinguishable, accessible, and linked to the affected surface and policy version.

CUSTOMER FACTUAL CONFIRMATION REQUIRED

Current scan workspace

URL Scan to Scope Matrix to GitHub findings.

The report keeps public observations, technical verification gaps, and counsel decisions separate. It never declares EU approval or guaranteed legal compliance.

AI surfaces1

Code findings0

Fix kits4

Evidence readiness14%

Observed signals

Sample scan loaded

Limits of this scan

Public scanning cannot determine contractual role, private runtime behavior, or counsel interpretation.

Technical receipt

HTTP: not fetched
Content: unknown
Bytes / files: sample

Shareable artifact

EU AI Act Article 50 Implementation Report

Every scan creates a customer-ready report URL with observations, likely obligations, confidence, source versions, technical recommendations, and counsel-review states.

EU AI Act Article 50Transparency Controls

0 / 5implemented

Last checked 06 Sept 2026, 20:31View evidence at isgenai.com/report/example-ai-product-com

Open report Trust RecordBadge SVG

EU AI Act Article 50(1)

Direct interaction disclosure

Requires legal interpretationCounsel reviewmedium risk

Add a runtime disclosure component attached to every direct AI interaction surface, with locale and accessibility variants.

EU AI Act Article 50(2)

Machine-readable marking of generated or manipulated content

Unable to verify externallyOut of scopelow risk

Use a marking adapter strategy: metadata or C2PA where appropriate, paired with signed evidence and a workflow-specific robustness suite.

EU AI Act Article 50(3)

Emotion recognition and biometric categorisation notice

Unable to verify externallyOut of scopelow risk

Maintain an exposed-person notice at first exposure to the operation of the system, linked to a privacy processing record and counsel-approved exception path.

EU AI Act Article 50(4)

Deepfake and public-interest content labelling

Unable to verify externallyOut of scopemedium risk

Route relevant releases through a labelled publication workflow with reviewer sign-off and durable provenance.

EU AI Act Article 50(5)

Accessible and machine-readable labelling context

Requires legal interpretationCounsel reviewmedium risk

Maintain a label copy registry, per-surface policy bindings, and event-level evidence export.

AI Surface Inventory

Canonical map of EU AI Act Article 50 exposure.

Each AI-facing surface carries owner, modality, role, user population, publication status, EU AI Act Article 50 mapping, implementation status, and evidence state.

mixed / unknown

Unclassified AI surface candidate

Counsel review

Owner: Needs assignment
Role: unknown
User population: unknown
Evidence: Unable to verify externally

EU AI Act Article 50(1), EU AI Act Article 50(2), EU AI Act Article 50(3), EU AI Act Article 50(5)

Not verified

GitHub repository scanner

Specific code findings, not questionnaire prose.

Repository scans map provider SDKs, generation routes, output pipelines, labels, metadata, publishing paths, and transparency logic to files and lines.

GitHub

No file-level findings yet

Unable to verify externally

Connect a public repository or run the server with GitHub CLI/token access for private repositories.

Least-privilege read access is enough for detection. Write access is only needed to open fix PRs.

Generate/Fix PR

Every finding gets an implementation path.

Fix kits include code, configuration, and review records. Draft PR creation works when a write-capable GitHub token is available and PR creation is explicitly enabled on the server.

EU AI Act Article 50(1) / minutes

Add EU AI Act Article 50 first-interaction disclosure

Production-ready accessible disclosure component for direct human-AI interaction, with policy version and evidence hook props.

customer chatbot / direct AI interaction surface

create: src/components/Article50Disclosure.tsx

EU AI Act Article 50(2) / hours

Wrap generation with runtime evidence and marking configuration

Adds a TypeScript runtime wrapper that records policy decision, provenance node, sidecar metadata, and known marking limitations.

generation API route

create: isgenai.runtime.ts

EU AI Act Article 50 / minutes

Add Transparency CI report to pull requests

Runs repository signal detection on every pull request and uploads a machine-readable implementation report.

.github/workflows/isgenai.yml

configure: .github/workflows/isgenai.yml

EU AI Act Article 50(3)/(4) / requires review

Record counsel determination before disabling controls

Creates an auditable override record for contextual determinations such as human editorial responsibility or obvious AI context.

emotion-biometric/deepfake/public-interest/assistive-editing determinations

create: isgenai.counsel-review.json

Control plane, not a badge

Five engines for operational AI transparency.

The moat is the graph from policy to runtime event to provenance to evidence. A label alone is too fragile for serious AI products.

Policy Engine

Versioned EU AI Act Article 50 rules become deterministic controls with source links, exceptions, and counsel annotations.

Provenance Graph

Every AI and human transformation becomes a reconstructable event, not a fragile one-bit AI-generated flag.

Marking Adapters

C2PA, metadata, visible labels, watermarking, and signatures stay swappable as the state of the art changes.

Transparency CI

Release checks test whether labels, metadata, and verification evidence survive real transformations.

Evidence Engine

Signed JSON, audit exports, and feasibility dossiers are produced from runtime events instead of screenshots.

Runtime Disclosure and Marking

TypeScript runtime, provenance graph, ledger, and fail-mode controls.

The launch SDK wraps generation, evaluates policy, emits disclosure text, applies marking adapters, signs evidence, records latency, and supports fail-open, fail-closed, and human-approval modes.

AI-generated or manipulated content

Runtime event feed

AI generationPolicy evaluated18 ms

ProvenanceInput/output hash linked41 ms

MarkingSidecar and signature adapters62 ms

LedgerEvidence entry hash chainedready

Evidence artifacts

Policy decision envelopeJSON

Implementable

Provenance event chainSigned hash chain

Implementable

Feasibility dossierPDF dossier

Counsel review

Audit exportOSCAL-ready JSON

Implementable

Security and privacy

Repository accessRead-only scan by default

Source retentionNo source persisted by scanner

EvidenceHash-chained local ledger

Runtime outagePer-surface fail mode

Transparency CI

Robustness testing belongs in every release.

Pull requests can report preserved disclosures, missing marking configuration, and counsel-review paths before release.

TransformationDetectionEvidence lossGate

Original artifact verification100%0%pass

JPEG compression or text minification96%4%pass

Resize, crop, or layout conversion84%13%watch

Screenshot, copy-paste, or social repost68%28%watch

Heavy paraphrase or adversarial laundering42%49%fail

Counsel Review

Legal determinations become auditable policy overrides.

Counsel can approve applicability, reject a scanner inference, or request missing customer facts. Those states are exported in the audit packet.

EU AI Act Article 50(1)

EU AI Act Article 50(1): Direct human-AI interaction disclosure

The scan cannot determine direct interaction from public signals alone.

route inventory, label copy version, component version, screenshot, release approval

Default state: customer facts needed

EU AI Act Article 50(2)

EU AI Act Article 50(2): Machine-readable marking and detection

The Code-of-Practice path is not confirmed, so the report keeps both Code and alternative-measures evidence visible.

artifact hash, adapter version, verification result, known limitations, feasibility rationale

Default state: technical verification needed

EU AI Act Article 50(3)

EU AI Act Article 50(3): Emotion recognition and biometric categorisation notice

No emotion recognition or biometric categorisation fact is currently indicated.

workflow classification, notice copy version, exposure timestamp, DPIA or privacy record reference, exception approval

Default state: technical verification needed

EU AI Act Article 50(4)

EU AI Act Article 50(4): Deepfake and public-interest content labelling

Deepfake or public-interest publication facts are contextual and should be reviewed before runtime controls are disabled.

use-case classification, label copy version, publication timestamp, reviewer approval

Default state: technical verification needed

EU AI Act Article 50(5)

EU AI Act Article 50(5): Accessible and context-appropriate labels

Any human-facing disclosure or label should be clear, distinguishable, accessible, and linked to the affected surface and policy version.

accessibility check, copy registry, surface binding, policy link

Default state: customer facts needed

EU AI Act Article 50(1)

Direct interaction disclosure

Disclosure placement is not bound to the runtime interaction context

Screenshot evidence, component version, policy rule version, route inventory, and release approval.

Default state: needs review

EU AI Act Article 50(5)

Accessible and machine-readable labelling context

No policy-linked copy registry observed

Copy version, surface binding, accessibility check, policy link, and exportable JSON evidence.

Default state: needs review

Hosted Transparency Record

Share your transparency records.

Public facts explain the implemented controls. Private evidence can be shared with buyers, auditors, or counsel when the customer authorizes it.

Need help implementing this?

isgenai.com/trust/example-ai-product-comCounsel review

Public facts

EU AI Act Article 50 scope has been assessed from observable product and declared workflow signals.
Implementation states separate technical controls from legal interpretation.
Trust Record is not an EU certification and should not be presented as one.

Private evidence

Runtime policy decisions
Artifact hashes and provenance graph
Robustness CI results
Counsel review notes and approved exceptions

Safer status language

Never reduce EU AI Act Article 50 to a single badge.

isGenAI uses operational states so engineers, counsel, auditors, and buyers understand what has been implemented and what still needs interpretation.

Implementable

Implemented

Verified

Counsel review

Technical limit

Private workflows

Authenticated surfaces can be reviewed with authorized evidence.

The public scanner only observes pages it is allowed to fetch. Logged-in exports, customer portals, evidence packets and private policy screens are handled through this consent-based evidence queue.

Authorized private workflow evidence

Submit logged-in or customer-only evidence for review.

Use this when the public scanner cannot see a disclosure, export, C2PA manifest, policy screen or evidence record because it sits behind authentication.