Case study · Runtime recovery

When incident AI actions diverge, the decision still has to be explained

A security operations team faced conflicting records after an AI-assisted remediation attempt during an active incident. Without Recon, explaining who authorized what—and what actually happened—would have depended on manual log stitching.

Simulation
BeforeRisk eventPolicy holdProof retained

Illustrative ribbon — not a production screenshot or KPI chart.

01 · Before

What was happening before Recon?

Incident response ran on fragmented exports: SIEM alerts, chat transcripts, and agent logs that did not share a single correlation from attempted action to outcome. When leadership asked why a credential rotation fired, the team could describe intent—but not defend the decision with attributable proof.

  • AI-assisted remediation crossed team boundaries without a shared decision record
  • Workflow revisions left conflicting narratives about what the agent attempted
  • Reviewers reconstructed timelines manually before every leadership briefing

02 · Governed moment

What consequential AI action created risk?

During an active incident, an autonomous security agent attempted credential revocation and endpoint isolation across a production segment—a high-blast-radius action that could disrupt live services if the wrong scope executed.

  • The agent attempted cross-segment remediation without a durable authority check
  • Conflicting workflow versions made it unclear which policy version should apply
  • IR leadership needed an answer before the next customer-facing status update

03 · Recon intervention

What did Policy permit, hold, or escalate?

Recon evaluated the governed interaction at the boundary. Policy is business control—not an architecture diagram. The team's security Policy required SOC lead approval before high-impact remediation executed.

  • Policy held the remediation attempt pending explicit SOC lead approval
  • Recon returned a require_approval decision the workflow could enforce
  • Operators received a clear escalation path instead of silent execution

04 · Proof retained

What Trust Receipt/Ledger evidence remained afterward?

Every governed interaction produces inspectable proof. The Trust Receipt captured the attempted action, Policy decision, and authority context. The Ledger appended durable evidence tied to the same correlationId—so the team did not depend on stitched logs to explain the hold.

  • Trust Receipt — open the artifact and see why Recon held the remediation
  • Ledger — confirm the entry persists for retention and investigation export
  • One correlationId threads from attempted action → decision → receipt → ledger entry
require_approvalTrust ReceiptSample

Recon requires approval before this action proceeds.

correlationId
tacorr_8f3a2b1c4d5e6f708192a3b4c5d6e7f8
receiptId
a1b2c3d4e5f6789012345678901234567890abcdef1234567890abcdef123456
decision
require_approval
policyId
policy_soc_lead_approval_gate
actionType
cyber_remediation_access_action

Ledger

The Ledger is your durable evidence spine—every Trust Receipt links back to the same correlationId for retention and audit.

correlationId
tacorr_8f3a2b1c4d5e6f708192a3b4c5d6e7f8
ledger.entryId
b2c3d4e5f6789012345678901234567890abcdef1234567890abcdef1234567890
Confirm sample in Ledger →

05 · What changed

What materially changed for operator or business?

Outcome frame: Decision explained. IR leadership received a defensible account of what the agent attempted, which Policy applied, and why execution paused—without asking the team to learn internal runtime vocabulary or rebuild timelines from chat exports.

  • Leadership briefing used attributable proof instead of reconstructed narratives
  • Operators spent less time reconciling conflicting workflow versions post-incident
  • The team could show what was attempted and why it did not execute unchecked

Outcome frame

Decision explained

Simulation only—illustrative incident-response narrative with no real customer data, no production outage claim, and no measured ROI.