Prompt Injection
A support bot follows instructions embedded in a customer-uploaded file.
Review AI-specific risks before they reach production workflows.
AI applications introduce new security questions: which prompts are exposed to untrusted content, which tools agents can call, which outputs are trusted, and which teams own the risk.
Prompt Hijacking
An agent prioritizes a hostile instruction over the user's task.
Indirect Prompt Injection
A browser agent reads a page that asks it to reveal private workspace data.
Data Leakage
A chatbot response includes confidential account notes from an internal knowledge base.
Excessive Agency
An agent can modify records and send messages without human approval.
Insecure Tool Use
An agent triggers a ticketing or email action based on untrusted input.
Unsafe Output Handling
A generated command, query, or code block is executed without policy checks.
AI Supply Chain Exposure
A third-party model integration lacks owner, purpose, and review metadata.
Risk scoring table
Filter synthetic findings by severity, system type, owner, and status.
8 synthetic findings shown
| Risk area | Severity | System | Owner | Status | Finding |
|---|---|---|---|---|---|
| Prompt Injection | High | Support Assistant | AppSec | Open | |
| Prompt Hijacking | High | Autonomous Agent | AI Platform | In Review | |
| Indirect Prompt Injection | High | Browser Agent | Product Engineering | Open | |
| Data Leakage | High | RAG Chatbot | Security Operations | Mitigating | |
| Excessive Agency | Medium | Workflow Agent | Product Engineering | Open | |
| Insecure Tool Use | Medium | Ticketing Copilot | AI Platform | Planned | |
| Unsafe Output Handling | Medium | Code Assistant | AppSec | Mitigating | |
| AI Supply Chain Exposure | Low | Vendor Model | Governance | Planned |
Remediation checklist
Convert finding review into operational change.
- Document the fictional system owner and business use case.
- Review prompt boundaries, retrieval sources, and connected tools.
- Require human approval for sensitive agent actions.
- Retest after policy, prompt, or permission changes.
FAQ
AI risk review basics.
Is this showing real customer risk?
No. All findings, teams, owners, systems, and evidence on this site are fictional demo content.
Can filters be used by keyboard?
Yes. Filters, rows, the dialog, and reset controls are keyboard-accessible HTML controls.