Guard
Enforcement for
AI agents.
Route AI outputs through ZeroDrift before they reach users. Send each output to the enforcement API and act on the verdict. Clean responses go through. Violations are rewritten, blocked, or escalated.
Runtime Enforcement Flow
One check between your model and the outside world.
//A
Application
or Agent
//B
LLM
Provider
//C
ZeroDrift
Guard
//D
Safe
Output
//A
Application or Agent
Agent, app, workflow, or integration.
//B
LLM Provider
OpenAI, Anthropic, or other LLM.
//C
ZeroDrift Guard
Anchor enforces: pass, rewrite, block, or escalate. Every verdict logged.
//D
Safe Output
Delivered, blocked, or escalated.
What Guard Enforces
Policy-level controls at runtime.
//A
Compliance Rules
Regulations, disclosures, and communication standards.
//B
Company AI Policies
Approved use cases, agent behavior, and output limits.
//C
Security Controls
Restricted data, unsafe outputs, and high-risk instructions.
//D
Rewrites
Replace unsafe or non-compliant language.
//E
Blocks
Stop prohibited outputs before delivery.
//F
Evidence
Log decisions, rules, reasoning, and final output.
Example Verdict
A verdict your application can act on.
curl -X POST https://api.zerodrift.ai/api/v3/content/validate \
-H "x-api-key: YOUR_API_KEY" \
-H "Content-Type: application/json" \
-d '{ "body": "The fund has beaten the market every quarter.", "rulepack": "financial_services_client_communications" }'
# verdict: pass, rewrite, block, or escalate
{
"verdict": "rewrite",
"risk_level": "high",
"violations": [
{
"rule_cited": "FINRA Rule 2210",
"policy_match": "Promissory Performance Language",
"explanation": "The message implies guaranteed safety and performance.",
}
],
"suggested_fix": "The fund may be appropriate for certain investors depending on objectives, risk tolerance, and time horizon. Past performance does not guarantee future results.",
"evidence_id": "evd_9x2f81",
}
{
"verdict":"rewrite",
"risk_level":"high",
"rule_cited":"FINRA Rule 2210",
"policy_match":"Promissory Performance Language",
"original_output":"This fund is a safe way to generate strong returns.",
"safe_output":"This fund may be appropriate for certain investors depending on their objectives, risk tolerance, and time horizon. Past performance does not guarantee future results.",
"delivery_status":"safe_response_delivered",
"evidence_status":"logged"
}
Provider Compatibility
Works with the LLM stack your team already uses.
Guard works with outputs from any model provider: frontier APIs, internal models, and enterprise AI integrations. No rebuild required.
OpenAI
Anthropic
Other LLM providers
Internal
models