Skip to content

Tamper-Evident Audit Trail

Every analyst action (acknowledge, resolve, dismiss, false-positive, assign), every remediation, and every configuration change — with a field-level diff — plus logins, logouts, and denied access attempts is recorded with who, when, their role, and origin IP to a write-once (WORM) S3 store with Object Lock.

The application can append but never alter or erase records, so the chain of custody holds even if a component is compromised. The trail is queryable in Athena and surfaced in scheduled reports. Choose GOVERNANCE or fully immutable COMPLIANCE retention at deploy.

Where to review it

  • The per-threat Audit History section.
  • The scheduled report's Audit Activity table.
  • The saved Athena query "Audit activity (who did what)" — database <stack>_siem, table audit.

A threat's full action history (status changes, notes, remediations) can be pulled directly from the WORM audit archive even after the live record has aged out of the operational store — so "what was done about this?" stays answerable (advanced+).

What each record carries

  • Threat context. Every status change, assignment, and note also records the threat's severity, first-seen time, MITRE tactic, and the rules that fired, plus whether the action reopened a closed threat. Time-to-acknowledge and time-to-resolve can be computed straight from the audit table.
  • Bulk actions. Every record written by one bulk status or bulk assign shares a bulk_id, and items a bulk action could not apply (not found, not permitted) are recorded with the reason.
  • AI playbooks. Each playbook request records the model used, whether it was a deployment default or a Settings override, whether it came from cache, and the outcome — including requests that finished in the background or failed.
  • Configuration changes. Exclusion conditions and trusted IPs are recorded one condition at a time, so the trail shows exactly what was silenced. If a change record ever has to be shortened, it says so. Defaults the editor fills in when you save are not recorded as your changes.
  • What raised the threat. Each threat action records whether a detection rule fired (and which) or which source produced it — the behavioral baseline, a GuardDuty or Security Hub finding, an AWS Config rule, flow forensics, a metric anomaly, or the configuration integrity check.
  • Sign-in sessions. Sign-in and sign-out records share a session id, so sessions can be counted and timed; a session with no sign-out ended at the one-hour session limit. A user's first appearance and every role change (for example, viewer to admin) are recorded too, for access reviews. Failed sign-ins and MFA challenges are handled by Amazon Cognito before AI SIEM sees the request, so they appear in Cognito and CloudTrail, not here.
  • Denied actions. Every action a role is not permitted to take is recorded with who tried it, the endpoint and method, the target, and the role it required — so probing is visible, not silent.

Configuration and posture history

Every configuration save, and a daily snapshot, keeps the full configuration in the WORM store for the audit-retention period, so you can see exactly which settings were in force on any day, long after the configuration's own version history has expired. A daily snapshot of the Compliance Posture view is kept the same way, so posture can be trended over time. Each control in the posture snapshot also records how many detection-rule exclusions narrow it, whether it counts as covered only with exclusions in place, and how many open threats it has to review. Each snapshot says what its score measures (score_basis: coverage, meaning the data sources and rules are on, unaffected by open threats) and lists any frameworks you turned off, so a framework that is missing by choice isn't mistaken for a gap. Snapshots are kept separate from the audit table, so they never appear in audit queries or reports.

If any audit record or snapshot fails to reach the WORM store, the <stack>-worm-audit-failures alarm fires.

Analyst notes

As you work a threat, add free-text notes in the Audit History section. Notes are append-only and appear in the threat's history (and the audit trail) alongside status changes and remediations, so the investigation reads as a running record.

Retention and archive

The live threat list in DynamoDB — the one you triage — is kept bounded by a daily archive pass, so dashboard totals stay exact and the list stays fast after months of activity. A threat leaves the live list when it is handled (resolved / false-positive / dismissed) and older than a short grace window, when a handled threat passes the hard age cap, or — as a ceiling — when the live count runs high (handled first, then lowest-severity and oldest). Open threats are not archived for age: they stay until someone triages them, unless you turn off Keep open threats past the age cap. If the high-water mark ever forces the archive to remove an open threat, you are told by email and on the dashboard.

Every removal is accountable per threat. Before deleting anything, each archive run writes a tamper-evident removal manifest listing every threat it will remove, with the threat's type, how it was detected, why it is being removed and whether it was still open. The run's audit records share one run id and report how many were actually removed and any that could not be. If the mass-delete safety limit stops a run, nothing is deleted and the audit trail records it as blocked, with the reason and the share of threats it would have removed; a run stopped by the per-run continuation cap is recorded as halted. The forensic copy in the S3 datalake and the WORM audit store keep their own, longer retention — archiving only prunes the operational store, never the record of what happened. Tune the grace window, hard age cap, keep-open setting, and the live-count high/low-water marks under Settings → Threat Retention & Archive.

Response times

Each status change that first acknowledges a threat (moves it off Active) or first resolves it (Resolved, False positive or Dismissed) records the measured time on its audit record, in seconds from detection (time_to_action_s, time_to_resolve_s), queryable in the Athena audit table. Reopening a threat does not reset its original times. Optional per-severity targets you set under Settings → Response Targets are part of every configuration snapshot, so results can be shown against the targets in force at the time. See the dashboard.

Discovery for suite integrations

Other Perfware products in the same account (such as AI Compliance) find AI SIEM's evidence stores through SSM parameters under /aisiem/<StackName>/, read with ssm:GetParametersByPath. Check contract-version first: version 2 is a superset of version 1, and keys are only ever added. Ignore keys not listed below: the path can also hold parameters from optional add-ons (for example remediation/assume-role-arn, written by the delegated-remediation stack).

Key Since Value
contract-version 1 2
version 1 AI SIEM release
tier 2 deployed tier (display and support)
logprocessor-stack 1 Log Processor stack (shared VPC, OpenSearch, Athena workgroup)
evidence/audit-bucket 1 WORM audit bucket
evidence/audit-lock-mode 1 GOVERNANCE or COMPLIANCE
evidence/audit-retention-days 1 Object Lock retention, days
evidence/datalake-bucket 1 datalake bucket (threats cold copy, raw sources)
evidence/config-bucket 1 config bucket
evidence/config-key 2 config/siem-config.json (includes response targets under sla)
evidence/glue-database 1 Glue database
evidence/audit-table 2 Glue table over the audit bucket (audit)
evidence/threats-table 2 Glue table over the threats cold copy (threats)
evidence/athena-workgroup 1 Athena workgroup
evidence/state-table 1 DynamoDB state table
evidence/config-snapshot-prefix 2 snapshots/config/ in the audit bucket
evidence/posture-snapshot-prefix 2 snapshots/posture/ in the audit bucket
evidence/archive-manifest-prefix 2 snapshots/archive/ in the audit bucket (removal manifests)
evidence/encryption 2 JSON: {"s3":"SSE-S3","stateTable":"KMS-AWS-managed","kmsKeyArn":"none"}
integrity/config-seal-key 2 JSON: state-table key of the config integrity seal, {"pk":"CONFIGSEAL#current","sk":"SEAL","hashAttribute":"config_hash"}
network/lambda-sg-id 1 AI SIEM's Lambda security group (add 443 ingress to reach its VPC endpoints)

Same-account reads need no change on AI SIEM's side: the bucket policies only require TLS, and no evidence store uses a customer-managed KMS key. The reader's own role needs Athena, Glue GetTable / GetPartitions, S3 GetObject / ListBucket on the audit and datalake buckets, DynamoDB GetItem / Query on the state table, and ssm:GetParametersByPath on /aisiem/<StackName>/*. If the account uses Lake Formation permissions, Glue reads also need Lake Formation grants.