EU AI Act

The EU AI Act (Regulation 2024/1689) establishes a risk-based regulatory framework for artificial intelligence systems operating in the European Union. It classifies AI systems by risk level — from minimal risk to unacceptable risk — and imposes documentation, transparency, accuracy, and human oversight requirements that scale with risk classification. Crowdee's hybrid AI and crowd architecture has been designed with these requirements in mind, ensuring that the tooling necessary to demonstrate compliance is built into the platform rather than bolted on after the fact.

AI System Documentation

Article 11 of the EU AI Act requires providers of high-risk AI systems to establish and maintain technical documentation before the system is placed on the market or put into service. This documentation must cover the system's intended purpose, the development methodology, the training data used, performance metrics, and known limitations.

Crowdee maintains technical documentation for each of the 14 verification pipelines. This documentation covers: the AI model family used, the provenance and scope of any training or fine-tuning data, the structured output schemas that each pipeline stage validates against, the intended use cases and content categories for which each pipeline is designed, the known limitations and edge cases where performance may be reduced, and the prohibited uses that would fall outside the scope of the system's design.

This documentation is available on request to enterprise customers and to regulators as required. Contact sales@crowdee.ai to request a copy for due diligence, regulatory submission, or internal risk assessment purposes.

Human Oversight

Article 14 of the EU AI Act requires that high-risk AI systems be designed to allow natural persons to effectively oversee them during use, including the ability to intervene, override outputs, and stop the system's operation. This requirement reflects the EU AI Act's core principle that AI systems operating in high-risk domains must not make or implement final decisions autonomously without the possibility of human correction.

Crowdee's crowd stages are the primary human oversight mechanism built into the platform. For Tier 2 and Tier 3 pipelines, the pipeline is designed so that human workers and domain experts independently review the AI's findings before any verdict is issued. The crowd stage does not simply rubber-stamp the AI output — workers form their own independent assessment based on the content itself, and the consensus of those assessments is what drives the crowd stage verdict. The AI findings are provided as context but workers are not required to agree with them.

Additional control points are available at the run level. The run status transitions to awaiting_crowd when a crowd stage is active, providing a clear signal that human review is in progress. Runs can be cancelled before the crowd stage completes if the content has been withdrawn or the decision circumstances have changed. These cancel and control capabilities are available via both the Platform interface and the API:

DELETE https://api.crowdee.ai/v2/projects/{projectId}/verification-runs/{runId}
X-API-Key: crw_YOUR_API_KEY

Transparency (Article 13)

Article 13 requires that high-risk AI systems be designed and developed to ensure that their operation is sufficiently transparent to allow deployers to interpret and use the system's output appropriately. Providers must supply deployers with instructions for use that enable them to understand the system's capabilities and limitations.

Every Crowdee verification run exposes a complete transparency record. For each AI stage, the record includes the prompt context that was assembled and sent to the model, the model's raw output before validation, the validated structured output, and the stage-level verdict and confidence score. For each crowd stage, it includes the individual worker responses, the consensus calculation, and the IRA score. The synthesis stage exposes its reasoning in the human-readable explanation attached to the final verdict.

Confidence scores provide a quantified uncertainty measure on every verdict. A confidence score of 95 indicates a high-certainty finding with strong supporting evidence across multiple stages; a confidence score of 45 indicates that the evidence is mixed and the verdict should be treated with caution. Confidence scores are designed to prevent deployers from treating all verdicts as equally reliable and to support proportionate decision-making downstream.

Prohibited Use Cases

Crowdee verification pipelines must not be used for purposes that fall within the prohibited AI practices defined in Article 5 of the EU AI Act or that otherwise violate Crowdee's Terms of Service. Prohibited uses include:

• Social scoring of individuals — using Crowdee's identity or content verification capabilities to produce reputation scores or trust ratings that determine access to services, employment, or social participation
• Real-time remote biometric identification in public spaces — Crowdee is not designed or licensed for use in biometric surveillance systems
• Manipulation of human behaviour — AI systems that exploit psychological vulnerabilities or operate subliminally contrary to the interests of the individuals they target
• Profiling based on sensitive attributes — deriving inferences about political opinions, religious beliefs, sexual orientation, or other protected characteristics from content analysis

These restrictions apply regardless of the pipeline used and regardless of whether the content being processed contains personal data. A use case that would otherwise fall within a prohibited category does not become permissible because the AI analysis is supplemented by crowd review.

If you are uncertain whether a proposed use of Crowdee is consistent with these restrictions, contact legal@crowdee.ai before proceeding. Full Terms of Service are available at crowdee.ai/terms.