Skip to main content

New AWS Detections: Stolen Access Keys, Secret Harvesting and GuardDuty Findings

Support
Fluency Support

Between November 14 and 16, 2026, Fluency released 21 new AWS detections, plus 2 supporting context rules, and updated 2 existing ones. They cover stolen access keys, secret harvesting, persistence, logging tampering, resource abuse and Amazon GuardDuty findings. They are active for every customer that sends AWS CloudTrail to Fluency; the GuardDuty detections apply where the GuardDuty integration is connected.

  • Built from a real-world attack. The stolen-key detections come from an AWS account compromise we investigated, and catch 36 of the 38 attacker access keys in that incident.
  • Tested against Stratus Red Team. We replayed the public attack logs of 36 Stratus Red Team AWS techniques. Before these releases, 20 went undetected; now 3 do, all of them sharing a resource with one outside account.
TypeDetection update
ReleasedNovember 14 to 16, 2026
Applies toFluency instances that collect AWS CloudTrail; the GuardDuty detections need Amazon GuardDuty connected
Action requiredNone to enable

How we validated​

The CloudTrail detections were checked against real attack activity, and every detection was measured against 30 days of real customer AWS activity before release. The GuardDuty detections were tested on real and simulated GuardDuty findings at each severity.

Samples from a real-world attack. In an AWS compromise we investigated, the attacker tested stolen access keys one by one: a GetCallerIdentity call, then attempts against services the key could not reach. We built the stolen-key detections from that activity and replayed it with all account details removed. They catch 36 of the 38 attacker keys. Against 14 days of activity on our other CloudTrail customers, they raised 3 alerts, all on roles.

Stratus Red Team attack replay. Stratus Red Team is an open-source attack-emulation tool that publishes the CloudTrail logs of each technique it runs. We replayed 36 of its AWS techniques (311 events) through Fluency's CloudTrail parser and every AWS rule, before and after these releases. The replay also found a gap in an existing rule, which we fixed: refused EC2 calls were not recognized as access denied.

Replay outcomeBeforeAfter
Incident on every occurrence36
Alert on every occurrence37
Alert when the attack crosses a threshold02
Alert on a new identity or network18
Alert only on first use by an identity910
Not detected203

Noise measured on real accounts. For each detection we counted how often its activity occurs in 30 days of CloudTrail from our customers, and tuned out routine sources: AWS's own automation, deployment pipelines, Kubernetes node launches and instance agents.

New and updated detections​

All the detections below need AWS CloudTrail except the three GuardDuty detections, which need the GuardDuty integration. "Incident" means a single match opens an incident.

DetectionStatusSeverityAlerts when
AWS_Principal_Recon_Multi_ServiceNew (Nov 15)IncidentAn access key or role checks its identity (GetCallerIdentity) for the first time in 30 days, then within an hour is refused two or more new APIs across two or more services: someone is finding out what a stolen key can reach.
AWS_Principal_Recon_New_Caller_IdentityNew (Nov 15)HighThe same identity check, followed within an hour by one newly refused API.
AWS_Access_Keys_Burst_From_IPNew (Nov 14)HighMore than four IAM users' access keys are used from one address within an hour.
AWS_Account_Guardrail_BypassNew (Nov 15)IncidentAn account tries to leave its AWS Organization, or EC2 serial console access is turned on; refused attempts included.
AWS_Security_Monitoring_DisabledUpdated (Nov 15)IncidentA security control is turned off. Now also covers deleting or detaching Route 53 Resolver DNS query logging and deleting Amazon Bedrock model-invocation logging.
AWS_CloudTrail_Event_Selectors_ChangedNew (Nov 15)HighThe events a CloudTrail trail records are changed, a way to stop logging while the trail still looks enabled.
AWS_Serial_Console_SSH_Key_SentNew (Nov 15)HighAn SSH key is pushed to an instance's serial console.
AWS_Roles_Anywhere_Trust_CreatedNew (Nov 15)HighAn IAM Roles Anywhere trust anchor or profile is created, giving long-lived access from outside AWS.
AWS_Bedrock_Model_Access_EnabledNew (Nov 15)HighAmazon Bedrock foundation-model access is requested or granted, the first step of model abuse with stolen keys.
AWS_Accelerator_Instance_New_PrincipalNew (Nov 15)HighAn identity launches GPU, machine-learning accelerator, FPGA or bare-metal instances for the first time in 30 days (cryptomining, model abuse).
AWS_Unauthorized_API_CallUpdated (Nov 15)First useA refused API call. Now also recognizes EC2's refusals, such as a denied attempt to read an instance's Windows password.
AWS_Secrets_Manager_Fan_OutNew (Nov 16)HighA person (IAM user, root or SSO session) reads 10 or more distinct Secrets Manager secrets within an hour.
AWS_Secrets_Manager_Batch_Read_New_PrincipalNew (Nov 16)HighA person reads secrets in bulk (BatchGetSecretValue) for the first time in 30 days.
AWS_SSM_Parameter_Fan_OutNew (Nov 16)HighA person reads 10 or more distinct SSM parameters one at a time within an hour.
AWS_SSM_Parameters_Batch_Fan_OutNew (Nov 16)HighA person reads 10 or more distinct SSM parameters through batch reads within an hour.
AWS_SSM_Remote_Command_New_PrincipalNew (Nov 16)HighAn identity runs an SSM command or starts a Session Manager session for the first time in 30 days; AWS's own automation excluded.
AWS_Instance_Connect_Key_New_PrincipalNew (Nov 16)HighAn identity pushes an SSH key through EC2 Instance Connect for the first time in 30 days.
AWS_Lambda_Public_Invoke_PermissionNew (Nov 16)HighA Lambda function is opened to be invoked by anyone (*).
AWS_Lambda_Code_Updated_New_PrincipalNew (Nov 16)HighA function's code is replaced by an identity that has not updated it in 30 days; CloudFormation deployments excluded.
AWS_Federation_Token_IAM_UserNew (Nov 16)HighAn IAM user requests a federation token for the first time in 30 days, a way to keep access after the key is disabled.
AWS_GuardDuty_Critical_FindingNew (Nov 16)IncidentGuardDuty raises a critical finding (severity 9 to 10).
AWS_GuardDuty_High_FindingNew (Nov 16)HighGuardDuty raises a high finding (severity 7 to 8.9), including stolen instance credentials used from another AWS account.
AWS_GuardDuty_Medium_FindingNew (Nov 16)MediumGuardDuty raises a medium finding (severity 4 to 6.9).

Two context rules, AWS_Principal_Caller_Identity and AWS_Principal_Refused, record the activity the recon detections use and raise nothing themselves.

What you can do​

No action is needed to receive these detections: they are active for every customer sending CloudTrail to Fluency.

  1. Connect Amazon GuardDuty if you have it on. Fluency's own rules cannot reliably see stolen instance credentials used from another EC2 instance in the same region; GuardDuty can, and the new GuardDuty detections turn its findings into Fluency alerts. GuardDuty findings arrive about 25 to 40 minutes after the activity starts. See AWS GuardDuty.
  2. Send CloudTrail from every account and region, including management events from your AWS Organization's management account, so the logging-tampering and organization detections see every attempt. See AWS CloudTrail.
  3. Treat an incident from AWS_Principal_Recon_Multi_Service, AWS_Account_Guardrail_Bypass or AWS_Security_Monitoring_Disabled as urgent. Each is rare in normal operation, and in the incident behind AWS_Principal_Recon_Multi_Service the attacker was testing keys in bulk.

Still not covered​

From the Stratus Red Team replay: sharing an AMI, an RDS snapshot or an EBS volume with one outside account. We are working on these, and on reading SSM parameters by path.

The rules' IDs, match conditions, exclusions and triage steps are in the AWS Detection Update release notes. AWS account compromise with stolen access keys in Attack Coverage describes the attack and how Fluency catches it. GuardDuty's finding types and severities are listed in the GuardDuty User Guide.

Questions: contact your Fluency representative, or raise a support ticket.

Splash Summary

Fluency has released 21 new AWS detections and updated 2, covering stolen access keys, secret harvesting, persistence, logging tampering, resource abuse and Amazon GuardDuty findings. The stolen-key detections come from an AWS account compromise we investigated and catch 36 of the 38 attacker keys in it. Of 36 Stratus Red Team AWS attack techniques replayed, the ones Fluency did not detect fell from 20 to 3.

The detections are active for every customer that sends AWS CloudTrail to Fluency. No action is needed to turn them on; connect Amazon GuardDuty to turn its findings into Fluency alerts.