AI Agent Oversight Pack - SLA & Scope

Mercury Security

AI Agent Oversight Pack — SLA & Scope
(v1.0, 2025)

The AI Agent Oversight Pack provides annual governance assurance for customer-facing and internal AI agents. This document defines the service scope, monitoring cadence, and client responsibilities.

1. Service Scope

Covered Agent Types

Reception & call answering agents
Customer support & FAQ agents
Knowledge retrieval agents (internal or external)
Social media posting agents
Workflow automation agents (simple decision support)

Included Governance Controls

Human-in-the-Loop (HITL) escalation procedures
Logging & retention policy
Bias & safety testing (scenario-driven)
Incident & escalation playbook
Hosting & assurance documentation (third-party vetted)

2. Service Level Agreement (SLA)

Monitoring Cadence

Monthly: Bias & safety sampling (20–50 test prompts per agent).
Quarterly: Review of logs, HITL escalations, and refusal handling.
Annually: Full Oversight Report with findings and recommendations.

Response Targets

Critical failures (unsafe completions, data leakage): reported within 5 business days of discovery.
Non-critical findings: documented in quarterly report.

Evidence Delivery

Oversight evidence is compiled into an Oversight Pack, including logs, test results, and policy updates. Delivered annually in PDF/DOCX.

3. Client Responsibilities

Clients must:

Provide timely access to agents and logs (via secure portal or API).
Assign a product/data owner for remediation follow-up.
Respond to Mercury requests within 5 business days.
Implement agreed remediation actions (Mercury does not enforce changes).

4. Exclusions

The Oversight Pack does not cover:

Model retraining or fine-tuning.
Development of new guardrails or moderation policies.
Formal legal compliance opinions or filings (e.g., DPIAs).
Vendor management or contractual negotiations.

5. Escalation Path

Unsafe or prohibited completions → documented & escalated to client owner.
Repeat failures → flagged in Oversight Report with recommended remediation.
Material governance failures → Mercury reserves right to suspend Oversight certification until remediation is complete.

6. Fees

$10,000 per year, organization-wide, for up to 3 production agents.
Additional agents may be covered via add-ons (see Add-On Catalog).

✅ The Oversight Pack ensures that AI agents remain governed, logged, and auditable throughout their lifecycle — not just at the point of initial deployment.