Topic starter
27/08/2026 2:11 pm
TL;DR: Security operations need controlled autonomy, not brittle SOAR scripts or unchecked AI agents, according to Mate. The central lesson is that agentic systems must reason and pivot inside governed boundaries, because access review cadences assume decisions persist long enough to be manually validated.
Editorial analysis by NHI Mgmt Group, based on content published by Mate: “Introducing Gamebooks: Can you script the whole game?”.
Key questions
Q: How should security teams retain command over autonomous AI agents that act on their behalf?
A: Security teams should treat AI agents as governed actors, not perimeter threats.
Q: Why do unbounded AI agents create more operational risk than SOAR playbooks?
A: SOAR is brittle because it assumes the world stays close to the script, but unbounded agents are risky because they can improvise past boundaries.
Q: What breaks when security automation loses investigation context between incidents?
A: Teams lose the reasoning behind prior thresholds, exceptions, and verdicts, so each new case has to be rediscovered from scratch.
Practitioner guidance
- Define investigation intent before tool steps Express each investigation as required evidence, permitted pivots, escalation triggers, and stop conditions before mapping any API calls or scripts.
- Separate decision authority from execution authority Let the agent choose the next capability, but keep tool access, scopes, and rate limits inside constrained flows that cannot be silently expanded.
- Preserve analyst reasoning in shared context Store why a threshold, exception, or verdict was chosen so later investigations inherit the rationale instead of repeating it as an ad hoc decision.
Bottom line: Controlled autonomy is the governing principle that keeps AI agents useful in security operations without letting them improvise past approved boundaries.
Explore further
View Full Forum → | NHI Foundation Course → | Our Services → | Read the full analysis →
27/08/2026 2:12 pm
Controlled autonomy is the right security model for AI agents. The false choice is between brittle scripts and unconstrained agents. Security operations need a third model in which the agent can reason and pivot, but only within explicit boundaries that define intent, evidence, and escalation.
A few things that frame the scale:
- 70% of organisations grant AI systems more access than they would give a human employee performing the exact same job, according to the 2026 Infrastructure Identity Survey.
- 19% of organisations give AI systems dramatically more access than human employees, nearly one in five granting unrestricted privilege, according to the 2026 Infrastructure Identity Survey.
A question worth separating out:
Q: When should teams test an AI agent’s response logic before enabling production actions?
A: Test in shadow mode whenever the procedure can affect access, containment, or production systems, especially when the agent is new, the environment is changing, or the response path has not yet been validated against prior investigations. The goal is to prove boundary discipline before the agent can act for real.
👉 Read our full editorial: Controlled autonomy for AI agents in security operations