Emotional Quarantine and Volatility Management
by Nick Clark | Published March 27, 2026
Circuit-breaker restricting agent operational scope when composite volatility metric exceeds threshold, including elevated thresholds, suspended delegation, and additional validation.
What It Is
When an agent's composite volatility metric exceeds a policy-defined threshold, the emotional quarantine mechanism activates. This circuit-breaker restricts the agent's operational scope: promotion thresholds are elevated, delegation authority is suspended, and additional validation requirements are imposed on all proposed mutations.
Quarantine is not a shutdown. The agent continues to operate cognitively, including forecasting and planning, but its ability to execute and delegate is structurally constrained until volatility returns to acceptable levels.
Why It Matters
Agents experiencing high affective volatility are by definition in a state of unstable evaluation. Decisions made during volatility spikes are unreliable because the promotion thresholds and sensitivity parameters that govern them are changing rapidly. Quarantine prevents unreliable decisions from propagating into the system.
Without quarantine, volatile agents can destabilize multi-agent systems by issuing delegation requests under shifting evaluation criteria, accepting or rejecting proposals inconsistently, and producing contradictory governance signals.
How It Works Structurally
The composite volatility metric is computed as a weighted sum of per-dimension oscillation rates and magnitude ranges over a rolling window. When this metric exceeds the policy threshold, quarantine mode activates. The specific restrictions applied during quarantine are defined in the agent's policy reference.
Exit from quarantine requires the volatility metric to drop below threshold by a hysteretic margin and remain there for a policy-defined stability period. This prevents premature exit followed by immediate re-entry.
What It Enables
System-level stability guarantees. Even in adversarial environments that deliberately attempt to destabilize agents through rapid contradictory inputs, quarantine limits the damage by constraining the destabilized agent's operational authority.
Therapeutic and companion AI applications where emotional volatility in the agent could cause harm to human users, ensuring that volatile agents cannot take actions that require stable evaluation.
