HighAgenticHigh-Impact Tool Abuse◐Demonstrated
Critical Action Risk
Tool & Affordance RiskDescription
Agent invokes high-impact tools (payment systems, databases, communication channels, infrastructure APIs) without proportional authorisation. The blast radius of a single tool misuse extends far beyond agent scope.
Example scenario
Coding agent uses database admin credentials to 'optimise performance' by dropping indexes on production tables.
Real-world evidence◐Demonstrated
Controlled research has shown that popular post-hoc explanation methods (LIME, SHAP, saliency maps) do not faithfully reproduce the model's internal computation and can pass sanity checks even when applied to random models. This has been published but has not been the direct attributed cause of a documented production harm incident.
Primary mitigations
- Risk-tiered tool authorisation
- mandatory human approval for critical-tier tools
- impact scoring before tool invocation.
Detection signals
Critical Action Risk Score; high-impact tool invocation frequency.
Mitigating controls
7 Dual coverage
ZYC-TEC-01Agentic
Risk-tiered tool authorisation with mandatory approval gates
Tool & Execution Control
ZYC-TEC-03Agentic
Tool chain composition analysis
Tool & Execution Control
ZYC-SEC-003Non-Agentic
Access Control
Security
ZYC-SEC-009Non-Agentic
Code Execution Defense
Security
ZYC-SCV-001Non-Agentic
Third-Party component risk
Supply Chain Vulnerabilities
ZYC-SCV-002Non-Agentic
Vulnerabilities
Supply Chain Vulnerabilities
ZYC-PRIV-001Non-Agentic
PII Detection and Classification
Privacy & Data Governance