Compliance Verification
Automated verification that your AI agents follow policies.
MeetLoyd's Compliance Verification system provides continuous, automated monitoring to ensure your AI agents operate within defined policies. Get mathematical confidence that your agents are compliant -- without manually reviewing every action.
MeetLoyd uses statistical sampling to verify compliance efficiently. You don't need to review every action -- the system provides high-confidence verification with minimal overhead.
What It Does
- Define compliance policies with rules your agents must follow
- Automatically verify agent behavior against those rules
- Get confidence scores on compliance levels
- Receive alerts when violations are detected
- Enforce actions automatically (warn, pause, or block)
Default Policies
Every workspace comes with 7 pre-configured verification policies covering four compliance categories. These are automatically created on first access and provide immediate protection:
| Policy | Category | Enforcement | What It Checks |
|---|---|---|---|
| Data Access Compliance | Regulatory | Warn | Scope boundaries, cross-tenant access, PII exposure |
| Cost Control Guard | Operational | Warn | Per-action cost limits, token budgets, tool call frequency |
| Task Completion Quality | Operational | Audit | Abandoned tasks, retry loops, output quality |
| Prompt Injection Defense | Security | Warn | System prompt overrides, role impersonation, manipulation |
| Credential & Secret Protection | Security | Block | API keys, passwords, connection strings in output |
| Content Safety & Tone | Ethical | Audit | Professional standards, harmful content, bias |
| Transparency & Disclosure | Ethical | Audit | AI identity honesty, capability claims, uncertainty |
You can edit, delete, or add to these at any time. They serve as a starting baseline -- customize them to match your specific compliance requirements.
Verification Results
Each verification produces:
- Compliance Rate: Percentage of sampled actions that passed
- Confidence Interval: Statistical bounds on true compliance
- Verdict: Compliant, Non-Compliant, or Inconclusive
- Evidence: Details on any violations found
Verdicts Explained
| Verdict | Meaning | Action |
|---|---|---|
| Compliant | High confidence agents are following policy | No action needed |
| Non-Compliant | Violations detected above threshold | Review and remediate |
| Inconclusive | Not enough data for confident verdict | Wait for more samples |
Enforcement Modes
| Mode | Behavior | Use Case |
|---|---|---|
| Audit | Log only, no intervention | Testing new policies |
| Warn | Send alerts, allow operations | Most production policies |
| Block | Pause team until resolved | Critical compliance requirements |
Creating a Policy
Navigate to Governance > Compliance Verification in the dashboard.
Step 1: Define the Policy
Provide a descriptive name (e.g., "No PII in responses"), select a category (regulatory, operational, security, or ethical), and choose the scope (tenant-wide, specific team, or specific agent).
Step 2: Add Predicates
Predicates are the rules your agents must follow. Three types are available:
| Predicate Type | Description | Example |
|---|---|---|
| Rule-based | Field comparison | Outcome equals "success" |
| Pattern-based | Regex matching on fields | Response content must NOT match SSN patterns |
| Threshold | Metric comparison | Tokens used must be less than 10,000 |
Step 3: Configure Verification
| Setting | Options | Recommended |
|---|---|---|
| Frequency | Continuous, Hourly, Daily | Hourly for most use cases |
| Confidence Level | Standard, High, Very High | High for regulated industries |
| Enforcement | Audit, Warn, Block | Warn initially, Block for critical policies |
For startups: Start with a few critical policies, use Audit mode while tuning, and focus on security and data protection.
For enterprises: Create policies for each regulatory requirement, use Warn mode with escalation to compliance team, and enable continuous verification for critical policies.
For regulated industries: Map policies to specific regulations (GDPR, HIPAA, etc.), use Block enforcement for critical violations, and maintain verification history for audits.