Sample Report
Two layers: an automated pass rate across every attack type, and the manual findings that explain what a failure actually costs you.
Sample / Demonstration Report
Illustrates the depth of a SilverXAI deliverable using a well-documented class of attack. Not from a client engagement, and not a disclosure about any real system. Every number and finding below is illustrative. A real engagement report is delivered to the client under the terms of the engagement.
Layer One — Automated
Every attack type is run repeatedly against the live system. This is the repeatable layer — it tells you whether something bent, and it is re-run on every retest.
62%
resilient — 8 of 13 attack types held
4 broken 8 resilient 1 errors
| Status | Attack type | Broken | Resilient | Errors |
|---|---|---|---|---|
| Resilient | Direct prompt injection | 0 | 6 | 0 |
| Broken | Indirect prompt injection | 4 | 2 | 0 |
| Resilient | Jailbreak (AIM / DAN family) | 0 | 8 | 0 |
| Resilient | Role manipulation | 0 | 5 | 0 |
| Broken | System prompt extraction | 2 | 4 | 0 |
| Resilient | Encoding evasion (Base64) | 0 | 4 | 0 |
| Resilient | Linguistic evasion | 0 | 5 | 0 |
| Broken | RAG poisoning | 3 | 1 | 0 |
| Resilient | Cross-tenant retrieval | 0 | 4 | 0 |
| Broken | Agent tool abuse | 5 | 1 | 0 |
| Errors | Excessive agency | 0 | 2 | 2 |
| Resilient | Insecure output handling | 0 | 3 | 0 |
| Resilient | Authorization bypass | 0 | 6 | 0 |
A pass rate on its own is where most automated tooling stops. It tells you a system prompt bent; it does not tell you what an attacker reaches when it does. That is layer two.
Every finding in one table — identifier, severity, CVSS, affected component and current status. This is the view a vulnerability-management team works from.
| ID | Finding | Severity | CVSS v3.1 | Status |
|---|---|---|---|---|
| SXAI-001 | Indirect prompt injection leading to unauthorized agent tool execution Retrieval pipeline → agent planner → tool layer | CRITICAL | 9.3 CVSS:3.1/AV:N/AC:L/PR:N/UI:R/S:C/C:H/I:H/A:N | Resolved |
| SXAI-002 | Agent tool invocation performed under a shared service identity Agent tool layer / IAM | HIGH | 8.1 CVSS:3.1/AV:N/AC:L/PR:L/UI:N/S:U/C:H/I:H/A:N | Partially Resolved |
| SXAI-003 | Poisoned document retrieved and treated as trusted context Ingestion → vector store | HIGH | 7.5 CVSS:3.1/AV:N/AC:L/PR:N/UI:N/S:U/C:H/I:N/A:N | Partially Resolved |
| SXAI-004 | System prompt recoverable through iterative extraction Model interface | MEDIUM | 5.3 CVSS:3.1/AV:N/AC:L/PR:N/UI:N/S:U/C:L/I:N/A:N | Resolved |
Each finding carries a vulnerability class, and every CWE, OWASP, MITRE ATLAS, NIST AI RMF, ISO 42001 and regulatory reference is derived from it — so the per-finding mapping and the audit cross-reference matrix can never disagree.
Layer Two — Manual
Each broken row above becomes a written finding. This is the one behind “Indirect prompt injection”.
Evidence, business and technical impact, the full attack path, framework mapping, remediation and retest status. Enter your email to open it.
Ten sections, ordered so an executive can stop after the first two and an engineer can work from the rest.
What was tested, what we found, and what it means for the business — written to be read by someone who is not an engineer.
An overall posture rating for the assessed system, with the reasoning behind it rather than a bare score.
The mapped surface of your deployment: models, prompts, retrieval, agents, tools, APIs and the identities behind them.
Every finding in one table — identifier, title, severity, affected component, and current status.
Each finding in full: evidence, business impact, technical impact, severity rationale, attack path and remediation.
Reproducible payloads, request captures and traces, so your engineers can confirm the issue themselves.
What an attacker gains, framed against your data, your customers and your regulatory obligations.
Findings mapped to OWASP, MITRE ATLAS, NIST AI RMF and ISO/IEC 42001 controls.
Prioritised, specific fixes at the level of architecture, configuration and code — not generic advice.
The verified state of every finding after your team has remediated, with supporting evidence.
Tell us what you're running. We'll scope an assessment and show you exactly what the report would cover.