Search the Atlas

Search risks, controls, and glossary terms

HighAgenticLack of Agent TransparencyTheoretical

Opaque Reasoning Chains

Accountability & Governance

Description

Stakeholders cannot inspect agent reasoning, tool invocations, or memory state. Explainability gap in autonomous multi-step decision making prevents effective oversight, debugging, and accountability.

Example scenario

Agent makes a consequential hiring recommendation with no explainable rationale, preventing HR from reviewing its basis.

Real-world evidenceTheoretical

While supply chain compromise is a Realized risk in software broadly (SolarWinds, XZ Utils), a confirmed incident in which an AI agent framework, plugin, or third-party model component was deliberately compromised and exploited in a production agentic deployment has not been publicly documented. The attack vector is plausible by analogy but undemonstrated for AI-specific agent supply chains.

No public incident on record — evidence level: Theoretical

Primary mitigations

  • Reasoning trace logging
  • step-by-step decision explanation
  • human-readable audit trails
  • interpretability tools integrated into agent pipeline.

Detection signals

Transparency Coverage Score; reasoning trace completeness rate.

Mitigating controls

5
Dual coverage

Related risks in Accountability & Governance