Attack in progressA leaked CI key of ‹ci-deploy-bot› is being used to break in: 6 accounts and machines created.Open the case
Agents
AI agent on: $3.78 of $15 spent today
Bedrock access for the lead and triage roles; hunt runs rules-only until its model is enabled.
Spend today $3.78 of $15 (25%)
| Role | Model |
|---|---|
| lead | anthropic opus class (us. profile) |
| triage | anthropic haiku class (us. profile) |
| refuter | amazon nova class |
Handled by Firekeeper 3
- Investigating
Investigating the case.
9 facts confirmed so far.
- Hunting
Watching your other account for the same keys and tooling.
Not seen there in the last 24 hours.
- Triaged
Cleared 31 routine alerts as normal activity.
Each matches your usual activity; all are still on Findings.
What the agent is doing live
- investigating
Found another key, ending 0003, made for ‹ci-deploy-b0t›.
A second look at the fake user listed its keys.
- hunting
Checked your other account for the same keys and tooling: not seen in 24 hours.
Cases like this often spread across accounts; this one has not, so far.
- automatic
Checked what the case's keys can reach: ‹prod-customer-exports› is readable.
Checked before anything is read, so you know what is at risk.
- investigating
Two networks: AS64501 (‹203.0.113.45›) and AS64509 (‹198.51.100.23›).
Neither network appears in your last 30 days.
- investigating
Tied every call to one tool fingerprint.
The same tool signature and machine-pace timing on 212 calls.
- investigating
Spotted ‹ci-deploy-b0t›, a fake copy of ‹ci-deploy-bot›.
One character differs; easy to miss in a list.
- investigating
Opened a case and started investigating.
A CI key created a user from outside its usual network.
- triage
Flagged it as urgent and handed it to the lead agent.
Within 40 seconds the key listed the account and created a user.
- automatic
Spotted the CI key of ‹ci-deploy-bot› in use from a new network, AS64501.
In 30 days this key was only ever used from your CI.
Agents' settings
What the agents may do, as signed. Each Change opens the setting on the Settings page.
- Agents are on
- The agents read findings with you, on the models below, within the budget below. Change
- Models per role
- Lead:
anthropic.claude-opus-4-5-20251101-v1:0; Triage:anthropic.claude-haiku-4-5-20251001-v1:0; Hunt:anthropic.claude-sonnet-4-5-20250929-v1:0; Second opinion:openai.gpt-oss-120b-1:0Change - Agents' day budget
- $15 a day Change
- May propose
- suppressions, hardening fixes, hunt leads Change
- Standing instructions
- None Change