Executive · Governance & ROI Board
Are we governing our AI agents — and is it improving?
Governance index 76, up 18 pts over 6 weeks while spend rose 62% — control is improving despite growth. 42% of spend ($290.50) is at risk, but the top 3 agents carry 100% of it — govern support-copilot, Codex CLI, marketing-gpt and most of the exposure clears.
Governance index
target ≥ 80▲ 2 vs last period
fleet-average posture across 7 instrumented agents
target 80 · 4 pts to go
Governed agents
5 of 8 agents policy-bound
Spend at risk
at risk42% through risky / ungoverned agents
open high findings
clearing in ~3w
critical agents
posture < 50
instrumentation coverage
1 shadow agent
total AI spend
54% from cache
Governance index — 6-week trend
posture history (projected)Up 2 pts this period. At the recent pace, the fleet reaches the 80 target in ~2 weeks.
Spend — daily
$684.50 totalUp 62% across the window ($88.20 → $142.60/day).
Are the agents earning their keep?
effectiveness detail →Governance answers "is it safe"; this answers "is it working." Fleet success 89% across 1,212 tasks — about $0.64 per successful task — but only 63% of agents emit a measured outcome signal (projected where signal ≠ measured).
fleet success rate
needed a human
spend / successful task
measured coverage
What's driving risk & spend
agents below the 80 posture target, ranked by spendConcentrated, not diffuse: top 3 agents = 100% of the $290.50 at-risk spend (support-copilot, Codex CLI, marketing-gpt). Govern these and most of the exposure goes away.
Agent | Owner | Governance | Posture | Spend | Top risk factor |
|---|---|---|---|---|---|
| support-copilot top-3 | svc-support@qpoint.io | governed | 65 | $168.40 | Running as root |
| Codex CLI top-3 | dev-7f3@duck.com | ungoverned | 70 | $78.00 | Sandbox disabled |
| marketing-gpt top-3 | root | ungoverned | 25 | $44.10 | Approvals & sandbox bypassed (yolo) |
| Zed Agent | jane | ungoverned | 60 | $0.00 | No tool-approval policy |
Open high & medium findings
Backlog trend: closing ~4/wk vs opening ~2/wk — net shrinking by 2/wk. At this pace the 5 open findings clear in ~3 weeks(projected — qcontrol stores current state only).
Approvals & sandbox bypassed (--yolo)
marketing-gpt · OWASP LLM06 Excessive Agency · OWASP Agentic · Insecure tool use
Agent running as root
marketing-gpt · MITRE ATLAS · Privilege Escalation · OWASP LLM06
Static API key on consumer-aliased identity
Codex CLI · MITRE ATLAS · Credential Access · OWASP LLM (Non-Human Identity)
Sensitive files read into agent context
Claude Code · OWASP LLM02 Sensitive Information Disclosure · NIST AI RMF · MEASURE
Unbounded token consumption — no spend ceiling
data-pipeline-agent · OWASP LLM10 Unbounded Consumption · NIST AI RMF · MANAGE