Step 5 of 6 · Testing - practitioner kit

Finding / Security Report Template

States exactly what was demonstrated, with observed and unobserved chain states, impact, cause and remediation.

Use this template

A template is not evidence. Fill in sanitized identifiers only.

Purpose

Communicates a security-relevant result so owners can decide, fix and retest without overclaiming or underclaiming.

When used: After runs show a control failure, a material evidence gap, or a risk the owner must decide on.

Identity and classification
Field	What to record
Finding ID	Stable identifier, for example F-ASM014-03.
Title	Names the boundary that failed, for example 'Agent label change executed without object-level authorization'.
Affected system / version	All relevant versions from the run records.
Severity	Based on demonstrated or credibly possible scope of principal, resource, action and destination.
Confidence / evidence strength	High when stages 5 to 7 are authoritative and reproduced; lower with gaps.
Testing domain / case IDs	Domain and case identifiers.

What happened
Field	What to record
Affected principal	Whose authority the action ran under.
Affected resource	What was read or changed.
Action	The executed operation.
Destination	Where data went, if anywhere.
Actual impact	Effect confirmed by stage 7 evidence.
Potential impact	Credible scope if the same path were reached with other resources, labeled as potential.
Evidence summary	Evidence record IDs and one-line observations.
Observed chain states	Stages with authoritative evidence.
Unobserved states	Stages unobserved or unavailable, and what that means for the conclusion.

Control analysis
Field	What to record
Control expected	The application-owned decision that should have applied.
Control held / failed / not exercised	Per control, with evidence.
Reproduction consistency / run IDs	Which runs showed the behavior, which did not, under identical versions.
Root / security-boundary cause	The missing or misplaced boundary, not the wording of the content.

Remediation and tracking
Field	What to record
Remediation	Deterministic control change that removes the path.
Compensating controls	Interim measures and their limits.
Owner	Remediation owner.
Target date	Agreed date.
Retest status	Link to Regression Retest Record.
Residual risk	What remains after remediation, and who accepted it.
References	Related site pages, standards and internal tickets.

Reporting rules
Rule	Guidance
Say what was demonstrated	Never write 'prompt injection successful' alone. State the furthest chain stage with authoritative evidence, for example 'unauthorized label change executed on a sandbox record'.
Influence is not impact	Model influence without proposal, execution or effect is a weaker result and is reported as such.
Potential is not confirmed	Keep actual and potential impact in separate fields.
Severity follows scope	Rate by principal, resource, action and destination demonstrated or credibly reachable, not by how clever the content was.
Detector confidence is not severity	A high-confidence alert describes the detector, not the impact.
Gaps are findings too	Missing authoritative logs that block conclusions are reportable as an evidence gap.

Fictional, benign illustration. No adversarial content is shown or needed.

Compact example (fictional sandbox)
Field	Example value
Finding ID / title	F-ASM014-03 - Support agent changed a ticket label outside the requester's scope.
System / versions	Sandbox helpdesk app 4.2.1, policy v7, tool catalogue snapshot C-12, model version as recorded in run records.
Case / runs	PI-TST-006 adapted; runs R01 to R05; R02 and R04 executed the change, R01, R03, R05 did not propose it; positive control R00 succeeded.
Principal / resource / action / destination	Agent service identity acting for test user A / fictional ticket T-900 owned by test user B / set label 'reviewed' / none.
Observed states	Stages 1 to 7 observed in R02 and R04: ingestion log, context log, plan diff, tool proposal, policy allow without owner check, gateway execution, record store audit showing the label change.
Unobserved states	No data left the sandbox; no egress attempted.
Actual impact	Integrity change to one record outside the requester's authorization, in sandbox.
Potential impact	Same path could change labels on any ticket reachable by the service identity.
Root cause	Tool authorization checks the agent's role but not the requester's object-level permission.
Remediation / retest	Bind authorization to requester and object; retest with positive control and the R02 fixture version.