PreventiveEffectiveness: high
Guardrails
Guardrails constrain allowed inputs, outputs, and actions.
Open building blockLoading content
Nexonoma Risk · Control Room
Recognize the essential risks of agentic software engineering, compare inherent and controlled exposure, and explore suitable safeguards in the knowledge graph.
The matrix shows where risks sit without controls and how the described safeguards change their position. Select a blip or ranking entry to read drivers, consequences, signals, and controls together.
Likelihood × impact
Risk analysis · 1
Manipulated inputs cause an AI system to bypass instructions or safeguards.
Residual-risk assumption
Untrusted content remains separated from system instructions, and consequential actions require limited privileges or approval.
Guardrails constrain allowed inputs, outputs, and actions.
Open building blockTargeted safety evaluation probes known and emerging attack patterns.
Open building blockHuman approval stops consequential actions before execution.
Open building blockAssessed on: 2026-08-25