Notifications
Clear all
Topic starter
18/08/2026 2:51 pm
TL;DR: AI-speed attacks and defensive refusal bias are pushing security teams toward autonomous defense, according to Fiddler, while its Behavior, Identity, and Control framework shows where agent governance breaks down: model behavior, granted identity, or runtime control. The deeper issue is that agent identity must be ephemeral because machine-speed action collapses access windows and audit assumptions.
NHIMG editorial — based on content published by Fiddler: Why AI Defense Has to Become Autonomous
By the numbers:
- Models refused roughly 70% of the time when students asked whether something was malware, even in a blue-team setting.
- 47%, rly half of organizations, 47%, do not trace their agents at all today.
Questions worth separating out
Q: How should security teams govern AI agents that can take runtime response actions?
A: Treat them as privileged NHI workloads with explicit scope, short-lived authority, and full action logging.
Q: Why do AI agents change IAM and PAM assumptions?
A: AI agents change IAM and PAM assumptions because they can act continuously, use tools directly, and execute without the human pacing that traditional review cycles expect.
Q: What breaks when teams cannot trace what an AI agent did?
A: Governance breaks first, because teams lose the evidence needed to decide whether the issue was behavior, identity, or control.
Practitioner guidance
- Define agent access by task lifetime Set access start and end conditions around task completion, not around human work hours or review windows.
- Instrument replayable agent telemetry Capture prompts, tool calls, data touched, approvals, and escalation events so every agent action can be reconstructed after the fact.
- Separate model policy from identity policy Review whether a failure came from the model refusing, the agent being over-privileged, or the runtime environment lacking controls.
What's in the full article
Fiddler's full blog covers the operational detail this post intentionally leaves for the source:
- The full BIC framework discussion showing how behavior, identity, and control failures differ in practice.
- The defensive refusal bias evidence and the security research context behind the 2.7 times refusal finding.
- The runtime governance model for autonomous defense, including how teams should think about telemetry and escalation.
- The article's discussion of why smaller security teams should start with the control layer first.
👉 Read Fiddler's analysis of autonomous AI defense and agent governance →
Autonomous AI defense and agent identity: are controls keeping up?
Explore further
18/08/2026 3:35 pm
Autonomous defense forces identity governance to separate model capability from delegated authority. A model that can answer questions is not the same thing as a system that can act, and the article is right to keep behavior, identity, and control distinct. That distinction matters because enterprises too often conflate model quality with security posture. The practical conclusion is that governance must be built around what the agent can do with granted authority, not just what the model can generate.
A few things that frame the scale:
- 92% agree governing AI agents is critical to enterprise security, yet only 44% have implemented any policies to do so, according to AI Agents: The New Attack Surface report.
- Only 52% of companies can track and audit the data their AI agents access, leaving 48% with a complete blind spot for compliance and breach investigation.
A question worth separating out:
Q: Who should own AI agent identity governance in an enterprise?
A: AI agent identity governance should sit jointly with IAM, platform security, and application owners because the risk crosses the runtime, the proxy, and the receiving service. No single team can see the whole delegation chain unless identity context is preserved end to end.
👉 Read our full editorial: Autonomous AI defense exposes the governance gap in agent identity
