Red Flags in Agent Spend
Runaway spend doesn't announce itself — but it leaves tracks. These seven red flags are the ones the incident database keeps showing up.
The seven flags
1. Off-hours activity — purchases outside business hours (the $12,400 night was 2 AM).
2. Unknown merchants — vendors nobody in the org recognizes.
3. Retry spikes — the same operation firing repeatedly.
4. Velocity jumps — more transactions per minute than any human workflow.
5. Small-payment floods — many tiny charges that add up silently.
6. Tier escalation — usage suddenly crossing into pricier bands.
7. Category drift — an agent spending outside its job (research agent buying compute).
What to do
Each flag maps to a rule: time-of-day, allowlist, velocity limit, category rule, cap.
At a glance
| Red flag | Rule that stops it |
|---|---|
| Off-hours activity | Time-of-day rule |
| Unknown merchants | Merchant allowlist |
| Retry spikes | Velocity limit |
| Small-payment floods | Daily ceiling + velocity |
| Category drift | Category rule |
FAQ
Which flag is the most dangerous?
Unknown merchants — it means the agent is buying from someone you've never vetted.
Are these retroactive or live?
Both — the flags describe what to look for in logs, and the rules stop them live.
How do I spot them in sipi.bot?
The audit log shows every decision: time, merchant, amount, rule.
Related
Stop the next $12,400 night.
One API call (or MCP tool) in front of every agent transaction — APPROVED, BLOCKED, or FLAGGED, deterministic, ~5 ms, fully logged.
See plans — from $99/mo Try a live check