Enterprise AI Governance: SOC2 Required Control

Every AI agent in your org, accounted for

CISOs lose sleep over three questions: Which agents are running? What can they access? Who owns the incident? Pretense answers all three: enforcement, not just visibility.

The 80-point enterprise readiness gap: answered

Which agents are running?

Pretense auto-discovers every AI agent routing through your network. Enrolled agents are profiled; unregistered agents trigger an immediate alert.

What are they authorized to do?

Define per-agent policies: which codebases, which data classes, which actions. Pretense enforces at the request level, not just logs after the fact.

Who is accountable when one goes wrong?

Every agent has a named owner. Every incident auto-creates a Jira/Linear ticket with CISO escalation SLA. Audit trails are 7-year retention for SOC2.

Agent Inventory

All active AI agents (live view)

Live: updated every 30s
AgentStatusRiskMutationsSecrets BlockedOwnerLast Active

Claude Code (eng-team)

Anthropic

Protectedlow1,2847Platform Eng2 min ago

GitHub Copilot (IDE)

OpenAI

Protectedlow8923DevEx8 min ago

Cursor (product team)

Anthropic / OpenAI

Scanningmedium3411Product Eng23 min ago

Unregistered Agent

Unknown

Threat Detectedcritical00Unassigned47 min ago

Critical: Unregistered AI agent detected

An agent not enrolled in Pretense policy was detected making LLM API calls. All traffic is blocked until policy is assigned. Incident #INC-2847 created. Escalation sent to CISO.

Authorization Matrix

Define what each agent can and cannot touch

Traditional DLP tools log what happened. Pretense prevents it. Set per-agent policies at the data class level, and Pretense enforces them inline, before any request leaves your network.

  • Per-agent policies: allow/block by data class (PII, secrets, IP)
  • Per-repo rules: different sensitivity for frontend vs. payments code
  • Time-bound access: temporary elevated permissions with auto-expire
  • Emergency revoke: kill a compromised agent in < 60 seconds
Claude Code: Policy Editor
# Agent: claude-code@eng-team
allow_codebases:["apps/*", "packages/*"]
block_patterns:["*.env", "secrets/*"]
block_data_classes:["PII", "API_KEYS", "SSN"]
mutation_level:"aggressive"
owner:"platform-eng@company.com"
escalation_sla_min:15
audit_retention_years:7
Last updated: 2 days ago● Active

Incident Accountability

When something goes wrong, who owns it?

Pre-defined escalation paths so no one scrambles when a breach happens. SOC2 requires documented accountability. Pretense builds it in.

Incident ScenarioPrimary OwnerEscalation PathResponse SLA
Agent leaks proprietary code to APICISOLegal + Engineering VP< 15 min
Secret (API key) detected in promptSecurity EngCISO< 5 min
Unregistered agent detectedPlatform EngCISO< 30 min
Agent authorization violationComplianceLegal< 1 hour
High-volume data exfiltration attemptSecurity SOCCEO + LegalImmediate

Integrates with your existing security stack

SplunkMicrosoft SentinelElastic SIEMJiraLinearPagerDutySlackServiceNowOkta SSOAzure AD

Ready to answer your CISO audit?

Enterprise teams get full agent governance: inventory, authorization, accountability, SIEM integration in one deployment.

SOC2 Type II in progress • HIPAA ready • ITAR compatible • 7-year audit retention

Ask me anything