1-minute read
What a defender needs to know
A CyberStrikeAI server used stolen credentials from eight AWS identities in four accounts over 16 days, Elastic Security Labs found. One EC2 role made 56 DescribeInstanceAttribute requests for userData in 141 milliseconds. On October 2, another role decrypted an SSM parameter and sent SendCommand; the command’s execution and outcome need checking in SSM command history. After 14 quiet days, the source IP returned to test credentials. The server ran CyberStrikeAI; triage the stolen credentials while the identity of any LLM decision-maker remains unverified.
Search CloudTrail for 43[.]240[.]29[.]11 and the related infrastructure IPs 43[.]240[.]29[.]28 and 43[.]240[.]29[.]33, then pivot on the real principal, assumed-role session, source IP and API operation. Elastic observed AWS events only from the first IP. Connect the burst of userData reads with GetSecretValue, decrypted GetParameter and later credential checks across the full September 30–October 16 interval. A ten-minute alert window catches the burst but misses that return.
Rotate affected access keys and secrets, revoke sessions and replace compromised instance profiles or instances. Restrict ec2:DescribeInstanceAttribute for userData to roles that need it; move startup secrets into managed storage. Verify in fresh CloudTrail events that retired credentials stop succeeding and replacement roles use approved networks, then reconcile the unexpected SSM command with its targeted machines.