Why it matters
Wiz’s JINX-2401 investigation follows compromised IAM user keys through attempted Bedrock access and creation of additional users, keys and policies. In the described administrator-access case, organization service control policies blocked model agreements and invocation even after the attacker created new identities. The report provides CloudTrail events and campaign indicators, but VPN addresses or naming patterns alone do not prove compromise. This is a 2024 incident record, not evidence of a new campaign or successful inference in every affected account.
My takeaway: Correlate unusual model invocation with IAM user, key and policy creation. Revoke compromised credentials and remove attacker-created identities; validate organization-level restrictions against both existing and newly created principals.