CloudTrail Events
AWS CloudTrail event names, indexed by what each one means in an investigation.
AWS documents what every API call does. It does not say which of several hundred event names matter in an investigation, or which field answers your question. The last two columns here do.
Before you read the table
- Data events are off by default. No GetObject usually means nobody enabled S3 data events, not that nobody read the data.
- IAM, STS and CloudFront log to us-east-1 wherever the caller was. Pull that region too, or you miss the IAM half of the intrusion.
- A run of AccessDenied across many actions is an identity probing its own permissions. It is often the clearest signal in the trail.
- userIdentity.type decides how you read the rest. Root should be near zero in a mature account.
55 rows.
| Event | Source | Purpose | Investigative use |
|---|---|---|---|
| ConsoleLogin | signin.amazonaws.com | A sign-in to the AWS Management Console. | The single most-read CloudTrail event. Success or failure is in responseElements.ConsoleLogin, and whether MFA was satisfied is in additionalEventData.MFAUsed: a success with MFAUsed "No" on a human account is worth a question every time. |
| AssumeRole | sts.amazonaws.com | Exchanges a caller identity for temporary credentials belonging to an IAM role. | The pivot point of nearly every AWS intrusion timeline. requestParameters.roleArn is where the caller went; requestParameters.roleSessionName is an attacker-chosen string and is often the most distinctive artifact in the whole trail. Follow the resulting session by matching responseElements.assumedRoleUser.assumedRoleId in later events. |
| AssumeRoleWithSAML | sts.amazonaws.com | Returns temporary credentials for a role, authenticated by a SAML assertion from a federated IdP. | The join point between your IdP logs and your AWS logs. If the AWS side shows a session your IdP never issued, the trust relationship or the IdP signing material is the problem, not AWS. |
| AssumeRoleWithWebIdentity | sts.amazonaws.com | Returns temporary credentials for a role, authenticated by an OIDC token. | Common and legitimate for CI/CD (GitHub Actions) and workload identity. The risk is a role trust policy whose condition is too loose: a `sub` claim wildcard that matches any repository, not just yours. |
| GetCallerIdentity | sts.amazonaws.com | Returns the account, user ID and ARN of the calling identity. | Harmless in isolation and used constantly by tooling, but it is also the reflexive first command after obtaining credentials of unknown provenance. Weight it by what surrounds it: GetCallerIdentity followed within seconds by broad List*/Describe* calls from a new IP is a recognizable orientation burst. |
| GetSessionToken | sts.amazonaws.com | Returns temporary credentials for the calling IAM user, optionally satisfying an MFA condition. | Converts long-term access keys into a temporary session. Worth noting because the resulting session credentials are what appear in subsequent events, not the original key. |
| GetFederationToken | sts.amazonaws.com | Returns temporary credentials for a federated user name chosen by the caller. | The federated name is caller-supplied, so it is an attacker-controlled string that will appear throughout the subsequent trail. |
| CreateAccessKey | iam.amazonaws.com | Issues a new long-term access key pair for an IAM user. | One of the highest-value single events in CloudTrail. Read requestParameters.userName against the caller: an identity creating a key for *itself* is routine-ish; creating one for a different, more privileged user is a persistence and escalation move in one call. |
| CreateLoginProfile | iam.amazonaws.com | Attaches a console password to an IAM user that previously had API access only. | A service/automation user suddenly gaining a console password is almost never legitimate. Pairs with UpdateLoginProfile, which changes an existing one. |
| UpdateLoginProfile | iam.amazonaws.com | Changes an IAM user's console password. | Changing another user's password is an account takeover with a clear before/after in the log. Changing your own is ordinary hygiene, so compare the caller against the target before anything else. |
| GetSecretValue | secretsmanager.amazonaws.com | Retrieves the plaintext value of a stored secret. | Application identities read the same few secrets constantly; a human or a newly-created role reading many secrets in sequence is the shape that matters. Baseline per-identity, not globally. |
| GetParameter | ssm.amazonaws.com | Reads an SSM Parameter Store parameter, optionally decrypting a SecureString. | The commonly-forgotten second secret store. withDecryption=true on a SecureString is the meaningful variant. GetParameters (plural) fetches up to ten at once and is the more efficient bulk-read. |
| Decrypt | kms.amazonaws.com | Decrypts ciphertext using a KMS key. | Extremely high volume in normal operation, so useless as a standalone alert, but excellent corroboration. A spike tied to one identity, or Decrypt calls against a key that identity has never touched, is the readable version of this signal. |
| GetPasswordData | ec2.amazonaws.com | Retrieves the encrypted local Administrator password for a Windows EC2 instance. | Narrow, deliberate, and rarely part of routine automation. Someone is trying to get an interactive session on a Windows host. |
| AttachUserPolicy | iam.amazonaws.com | Attaches a managed policy to an IAM user. | Read requestParameters.policyArn first: `arn:aws:iam::aws:policy/AdministratorAccess` attached to anything is the end of the escalation story. AttachRolePolicy and AttachGroupPolicy are the same move against a role or a group. |
| PutUserPolicy | iam.amazonaws.com | Embeds an inline policy document directly on an IAM user. | The policy JSON is inline in requestParameters.policyDocument. Read it, do not just note the event. Inline policies do not appear in managed-policy inventories, which is precisely why they get used. PutRolePolicy and PutGroupPolicy behave identically. |
| CreatePolicyVersion | iam.amazonaws.com | Creates a new version of an existing customer-managed policy. | Classic and easily missed: the policy ARN attached to the principal never changes, only its contents. setAsDefault=true makes the new version live immediately. Any review that checks "which policies are attached" and not "what do they currently say" will miss this entirely. |
| SetDefaultPolicyVersion | iam.amazonaws.com | Switches which existing version of a managed policy is active. | Can activate a permissive version created long ago and left dormant, so the malicious change and its activation may be widely separated in time. Check the version's creation date, not just this event's. |
| UpdateAssumeRolePolicy | iam.amazonaws.com | Rewrites a role's trust policy: who is allowed to assume it. | Read the new document for a Principal naming an AWS account that is not yours. The role keeps its name and its permissions and simply becomes assumable by a stranger, which is what makes this a durable cross-account backdoor. |
| AddUserToGroup | iam.amazonaws.com | Adds an IAM user to a group, inheriting the group's policies. | Quieter than attaching a policy directly, and the resulting permissions are one indirection away from the log line. Resolve what the group actually grants before dismissing it. |
| CreateUser | iam.amazonaws.com | Creates a new IAM user. | In an environment with federated access, creating a *local* IAM user is a deliberate step around the IdP, and therefore around its MFA and conditional access. Look for the CreateAccessKey or CreateLoginProfile that almost always follows within seconds. |
| CreateRole | iam.amazonaws.com | Creates an IAM role with a trust policy. | Common in legitimate infrastructure-as-code, so judge it by the trust policy in requestParameters.assumeRolePolicyDocument: an external account principal on a brand-new role is the finding. |
| DeactivateMFADevice | iam.amazonaws.com | Removes an MFA device from an IAM user. | Rarely legitimate outside a help-desk workflow that should have its own ticket. Pairs with EnableMFADevice re-registering an attacker-controlled device immediately after. |
| EnableMFADevice | iam.amazonaws.com | Associates an MFA device with an IAM user. | Ordinary at onboarding. Suspicious immediately after DeactivateMFADevice, or on an account the legitimate owner did not touch. Compare against the help-desk record. |
| GetAccountAuthorizationDetails | iam.amazonaws.com | Returns every user, group, role and policy in the account in one response. | A single call that dumps the account's entire authorization model. Almost nothing legitimate needs it outside a scheduled audit tool you can name. Treat an unexplained occurrence as a mapped account. |
| ListBuckets | s3.amazonaws.com | Lists every S3 bucket in the account. | Cheap, common in tooling, and the natural first step before data theft. Its value is positional. What it precedes matters more than the call itself. |
| GetBucketPolicy | s3.amazonaws.com | Reads a bucket's resource policy. | Reading policies across many buckets in sequence is reconnaissance for a misconfiguration, not administration. GetBucketAcl and GetPublicAccessBlock are read the same way. |
| DescribeInstances | ec2.amazonaws.com | Lists EC2 instances and their configuration. | Very high volume from monitoring and orchestration. Meaningful only when the caller is new, the source IP is new, or it appears inside a burst alongside DescribeSecurityGroups, DescribeVpcs and DescribeSnapshots. |
| DescribeSecurityGroups | ec2.amazonaws.com | Lists security groups and their rules. | Reconnaissance for reachable services and for a group loose enough to abuse. Read together with AuthorizeSecurityGroupIngress, which is the follow-through. |
| ListSecrets | secretsmanager.amazonaws.com | Lists secret names and metadata, not their values. | The natural precursor to GetSecretValue. ListSecrets followed by a run of GetSecretValue calls is a complete credential-theft sequence in two event names. |
| DescribeSnapshots | ec2.amazonaws.com | Lists EBS snapshots. | Snapshots are whole-disk copies, so they are a data-theft target that bypasses every file-level control. Enumerating them is the step before sharing or copying one out. |
| ListUsers | iam.amazonaws.com | Lists the IAM users in the account. | Individually unremarkable. Read as part of an enumeration burst with ListRoles, ListPolicies and ListAccessKeys. The burst is the signal, not the line. |
| StopLogging | cloudtrail.amazonaws.com | Stops a CloudTrail trail from recording events. | About as close to a definitive finding as CloudTrail offers. The event is itself recorded, so it marks the exact moment visibility ended, and the gap that follows is evidence in its own right. Establish what the trail covered before assuming the blackout was total; a second trail or an Organization trail often kept running. |
| DeleteTrail | cloudtrail.amazonaws.com | Deletes a trail configuration. | Louder than StopLogging and just as conclusive. Events already delivered to S3 survive the trail's deletion: check the bucket before concluding the history is gone. |
| UpdateTrail | cloudtrail.amazonaws.com | Modifies a trail: its S3 destination, its multi-region flag, its log file validation. | The subtle sibling of StopLogging: the trail still exists and still reports as enabled, while quietly no longer covering the region the attacker is working in, or no longer validating its own log files. Diff the request against the previous configuration. |
| PutEventSelectors | cloudtrail.amazonaws.com | Changes which management and data events a trail records. | Surgical log tampering. The trail runs, but the specific event classes the attacker is about to generate are no longer selected. Read the selectors themselves; the event name alone tells you nothing. |
| DeleteFlowLogs | ec2.amazonaws.com | Deletes VPC flow logs. | Removes the network-layer record while CloudTrail keeps recording the control plane, so this event survives to describe the destruction of the other evidence. |
| DeleteDetector | guardduty.amazonaws.com | Deletes the GuardDuty detector, disabling threat detection for the account. | Unambiguous. Note the quieter variants that achieve the same outcome: CreateFilter with an archive action, and UpdateDetector disabling specific data sources. |
| StopConfigurationRecorder | config.amazonaws.com | Stops AWS Config recording resource configuration changes. | Removes the "what did this resource look like before?" record, which is what you would otherwise use to prove a policy or security-group change. Costs the investigation more than it costs the environment. |
| GetObject | s3.amazonaws.com | Reads an object from an S3 bucket. | Data event: NOT recorded unless the trail was explicitly configured for it. Absence proves nothing. When it is available it is the closest thing to direct proof of data being read. Volume and breadth, one identity, many keys, short window, is the shape to look for. |
| PutBucketPolicy | s3.amazonaws.com | Replaces a bucket's resource policy. | Read the policy document for `"Principal": "*"` or an unfamiliar account ARN. This is how a private bucket becomes readable by the internet or by a stranger's account, in one call. |
| DeletePublicAccessBlock | s3.amazonaws.com | Removes the account- or bucket-level guardrail that prevents public access. | Rarely legitimate and always deliberate. The block exists specifically to stop the next step. Expect PutBucketPolicy or PutBucketAcl immediately after. |
| ModifySnapshotAttribute | ec2.amazonaws.com | Changes who can access an EBS snapshot, including sharing it with another account or making it public. | A whole-disk copy handed to an account you do not control. Read requestParameters.userId for the recipient: an unfamiliar 12-digit account number here is the finding, and it is often the attacker's own. |
| ModifyDBSnapshotAttribute | rds.amazonaws.com | Changes who can restore an RDS snapshot, including sharing it with another account. | The database equivalent of the EBS snapshot share, and the cleanest way to steal a whole database. `all` in the values to add means public. |
| CreateSnapshot | ec2.amazonaws.com | Creates a point-in-time copy of an EBS volume. | Ubiquitous in backup automation, so judge by caller and by what follows. CreateSnapshot followed by ModifySnapshotAttribute or CopySnapshot to another region is a staging-then-exfiltration pair. |
| RunInstances | ec2.amazonaws.com | Launches EC2 instances. | Judge by shape, not by presence: unusual instance types (GPU/compute-heavy), unusual regions, and an IAM instance profile more privileged than the workload needs. userData is base64 in the request and is executed at boot. Decode it. |
| ModifyInstanceAttribute | ec2.amazonaws.com | Changes an instance attribute, including its userData script. | The userData variant is a code-execution primitive: rewrite the script, stop and start the instance, and it runs as root or SYSTEM at boot. Watch for the StopInstances/StartInstances pair that must follow for it to take effect. |
| SendCommand | ssm.amazonaws.com | Runs a command on managed instances through Systems Manager. | Direct remote execution on hosts from the AWS control plane, with no SSH, no open port, and no network-layer evidence. The command itself is in the request parameters. Read it. AWS-RunShellScript and AWS-RunPowerShellScript are the documents to look for. |
| StartSession | ssm.amazonaws.com | Opens an interactive Session Manager shell on an instance. | An interactive shell with no bastion and no inbound rule. Session content is only recoverable if session logging to S3 or CloudWatch was configured in advance. Check before assuming you can see what was typed. |
| UpdateFunctionCode | lambda.amazonaws.com | Replaces the deployed code of a Lambda function. | Backdoors an existing trusted function rather than creating a suspicious new one. The function keeps its name, its triggers and its execution role, so it keeps its permissions too. Compare against the deployment pipeline: a code update that did not come from CI is the finding. |
| CreateFunction | lambda.amazonaws.com | Creates a Lambda function with a role and optional triggers. | Read the execution role and the trigger. A function wired to a scheduled EventBridge rule is a cron job in the cloud, and one wired to a CloudTrail or Config event is a reaction to your own defensive tooling. |
| AuthorizeSecurityGroupIngress | ec2.amazonaws.com | Adds an inbound rule to a security group. | The finding is in the CIDR and the port: `0.0.0.0/0` on 22, 3389, or a database port is the classic. A single-address /32 is quieter and more likely to be the attacker's own egress IP, which makes it excellent pivot data. |
| CreateKeyPair | ec2.amazonaws.com | Creates an EC2 key pair; AWS returns the private key once and never again. | A key pair created during an incident window is attacker-held by definition, which AWS handed the only copy of the private key to whoever made the call. ImportKeyPair is the variant where the attacker supplies their own public key. |
| PutRule | events.amazonaws.com | Creates or updates an EventBridge rule: a schedule or an event pattern with targets. | A schedule expression here is a cloud-native cron. Paired with a Lambda target it is a self-restarting persistence mechanism that no host-based tool will ever see. |
| CreateStack | cloudformation.amazonaws.com | Creates a CloudFormation stack from a template. | One event that creates many resources. The individual resource creations appear separately with a CloudFormation service principal as the caller, which can make them look automated and routine. Trace back to the stack and read its template. |