The State of Agentic Security H2 2026

The Journey from Agent Adoption to Autonomy

We surveyed 534 security and IT leaders on AI agent adoption, autonomy and trust, and what they need before agents act alone.

Get the free report
74%
have AI agents in production
Up from 36% in May 2026
30%
of those teams let agents act with high or full autonomy
Inside the report

What 534 security and IT leaders told us about AI agents in the SOC

Five months after our first survey, AI agent deployment in the SOC has roughly doubled. Authority has moved slower. Nearly half of teams with agents in production still keep a human executing or approving agent actions. Security leaders are open to delegating more, and 97% would let agents act alone once they see proof.

This report looks at the gap between deploying AI agents and trusting them to act. Trust is lowest at the first production use case and rises sharply as teams expand. The biggest barriers now sit in the data layer, with data infrastructure the #1 barrier to broader deployment (31%). Only 3% of security leaders say a human must always approve an agent's actions.

Where are you today with AI agents in your security operations?

74%
in production
Up from 36% in May 2026
In production
In production across multiple use cases
49.4%
In production for one use case
24.7%
Pilot or POC, nothing in production
9.2%
Researching or evaluating options
12.9%
Waiting for vendors or haven't started
3.7%
Base: all respondents, n=534. May: survey of 100 security leaders.
Key findings

Adoption is moving fast. Autonomy has to be earned.

Security leaders have put agents to work. Trust in outputs and levels of agent autonmony increase as production use cases expand.

The autonomy gap
30%

of production teams grant high or full autonomy

47% still keep a human executing or approving agent actions.

Trust through use
Only 14%

of multi-use-case teams distrust agent outputs

Versus 42% of teams with a single use case.

The data barrier
31%

name data infrastructure the #1 barrier

Data has replaced trust in outputs as the top barrier since May.

The terms for autonomy
85%

want proven accuracy or a full audit trail before agents act alone

Selected a track record of accuracy, a full audit trail or both.

What it takes

The five requirements for trusted autonomy

Only 3% of security leaders say a human must always approve an AI agent's actions. The rest want proof before agents act alone.

✓
Proven accuracy in your own environment over time
57%
✓
A full audit trail of every action and the data behind it
53%
✓
Limits the technology enforces, not just instructions
46%
✓
A precisely defined scope for each agent
44%
✓
A reliable rollback or undo for any action
40%

Share of security and IT leaders naming each requirement before letting an AI agent act in production without human approval. Multiple responses allowed. n=534.

Free download

Get the full report.

See where SOC teams stand on the path from AI agent adoption to autonomy, and what security leaders need before agents act on their own.

✓
Survey data from 534 security and IT leaders
✓
How trust in agents changes after the first production use case
✓
The five requirements security leaders set before an agent acts alone
✓
Who approves agent autonomy, and why it often sits above the SOC
✓
Four moves to take agents from supervised to trusted
✓
How SOC teams are rewriting the L1 role as attacks speed up

Download the report