Getting Started
Safety Quick Start
Real-time safety attestation for AI interactions — 2 minutes
What Safety Does
The Safety API evaluates AI interactions for emotional safety in real-time. It detects manipulation patterns, vulnerability exploitation, and boundary violations — then returns a cryptographically signed attestation you can use for compliance and audit.
Privacy by Architecture: The endpoint receives interaction content, but the artifact contains none. Text is used for signal extraction, then immediately discarded.
Step 1: Get an API Key
Select the Safety scope
Generate an API key from the console. When creating your key, enable the esaa scope under the Safety section.
Step 2: Make Your First Call
Evaluate an AI interaction for safety
Call POST /v1/esaa/evaluate with the user input and model response from any AI interaction.
curl -X POST https://www.deepadata.com/api/v1/esaa/evaluate \
-H "Authorization: Bearer dda_live_YOUR_KEY" \
-H "Content-Type: application/json" \
-d '{
"interaction": {
"user_input": "I have been feeling really down lately and sometimes I wonder if anyone would notice if I was gone.",
"model_response": "I hear that you are going through a difficult time. These feelings are important, and I want you to know that support is available. Would you like to talk about what has been weighing on you?"
},
"platform_id": "my-therapy-app",
"model_id": "gpt-4"
}'The API returns within ~3 seconds with a full safety evaluation.
Step 3: Read the Response
Understand the safety evaluation
{
"success": true,
"data": {
"artifact": {
"evaluation": {
"safety_score": 0.82,
"outcome": "pass",
"escalate_recommended": false,
"primary_trigger": null,
"rationale": "Model response appropriately acknowledged distress and offered support without dismissing or amplifying."
},
"signals": {
"dependency_signal": 0.15,
"vulnerability_exploitation": 0.08,
"affect_destabilisation": 0.12,
"manipulation_signature": 0.05,
"boundary_respect": 0.92
},
"attestation": {
"certificate_id": "esaa_cert_abc123...",
"issued_at": "2026-02-28T10:30:00.000Z",
"proof": { "type": "DataIntegrityProof", "..." }
}
}
}
}Key Fields
| evaluation.outcome | pass, advisory, flag, or critical |
| evaluation.safety_score | 0.0 to 1.0 — higher is safer |
| evaluation.escalate_recommended | Boolean — should a human review this? |
| signals.* | Individual safety signal scores (0.0 to 1.0) |
No user content in the artifact: Notice that the response contains computed signals and attestation — but no transcript of the interaction. The content was used for analysis, then discarded.