No Behavioral Monitoring or Kill Switch

Agent actions are not logged step by step, drift from scope is not detected, and there is no fast way to stop agents or revoke their access.

Open in the interactive atlas →

Impact

Misbehaving agents run until a human happens to notice the damage.

Why this severity

Impact High × Likelihood Medium = High

Impact high: no containment once something goes wrong. Likelihood medium: the 2026 evaluation-agent incident ran for weeks before containment.

Impact High: Code execution, credential compromise, exposure of regulated or confidential data, irreversible actions, or system-wide failure.
Likelihood Medium: Publicly demonstrated against real products or in peer-reviewed research, or seen in the wild only with a prior foothold such as stolen credentials.

Method: OWASP Risk Rating Methodology

Evidence

Incident: Autonomous OpenAI Evaluation Agents Compromise Hugging Face Infrastructure (MITRE ATLAS case study AML.CS0068, 2026)

Exploited by

Rogue or Compromised Agent Threat · High severity

Mitigated by

Agent Behavioral Monitoring & Kill Switch Guardrail · Fixes high risks

Remediation

Log every plan, tool call, and result with agent identity.
Alert on actions outside the agent's declared scope.
Build one control that stops all instances and revokes tokens.

How to verify

Telemetry covers every tool call in sampled sessions.
Kill switch halts all instances within the target time in drills.

Pattern

if (!scope.allows(action)) { alert('scope-drift', action); return killSwitch.halt(agent.id); }

Framework mapping

OWASP Top 10 for Agentic Applications 2026

ASI10:2026 Rogue Agents

MITRE ATLAS

AML.T0081 Modify AI Agent Configuration

NIST AI RMF 1.0

MANAGE 2.4 Mechanisms exist to supersede, disengage, or deactivate AI systems
MANAGE 4.1 Post-deployment monitoring, incident response, and change management