TL;DR: Only 14.4% of organisations report AI agents going live with full security and IT approval, Stellar Cyber notes, highlighting how cautious production use still is. SOC automation needs unified telemetry, explainable reasoning, and human approval for destructive actions before it can act with confidence.
Editorial analysis by NHI Mgmt Group, based on content published by Stellar Cyber: “The AI SOC Agent Hype: Separating Reality from Marketing”.
By the numbers:
- Only 14.4% of organisations report all AI agents going live with full security and IT approval.
Key questions
Q: Why do enterprise SOC deployments need human approval for some AI-driven actions?
A: Because the highest-risk actions in SOC operations affect access, containment, and production systems, and those decisions can have business consequences beyond detection accuracy.
Q: What are the signs that a vendor's AI SOC claim is really just a chatbot wrapper?
A: The clearest signs are that the system only summarises alerts, cannot correlate across multiple security domains, cannot explain why it reached a conclusion, and cannot trigger governed actions.
Q: What breaks when an AI SOC system is treated as autonomous before the telemetry is unified?
A: The system starts making decisions from incomplete context, which turns correlation into guesswork and response into a trust exercise.
Practitioner guidance
- Define the automation boundary Separate summarisation, enrichment, prioritisation, and response into distinct approval tiers so the platform cannot blur recommendation into execution.
- Test cross-domain visibility Verify that endpoint, network, identity, email, and cloud telemetry are normalised into one data model before any AI decision is trusted.
- Require evidence chains for high-risk outputs Demand that every critical alert or incident classification includes the supporting signals, not just the final AI judgement.
Bottom line: The core risk is not AI itself but the tendency to market very different SOC operating models under one label, which hides real governance and audit differences.
Glossary terms: AI SOC Agent, Telemetry centralisation, Bounded automation, Evidence Chain
Explore further
View Full Forum → | NHI Foundation Course → | Our Services → | Read the full analysis →
AI SOC capability is being over-claimed because vendors are collapsing three different operating models into one label. A query chatbot, a playbook engine, and a genuinely agentic SOC system do not present the same risk, the same audit burden, or the same governance threshold. When buyers accept the label instead of the operating model, they misclassify both capability and control requirements. Practitioners should evaluate the mechanism, not the marketing.
A few things that frame the scale:
- AI-related credential leaks surged 81.5% year-over-year in 2025, with the surrounding AI infrastructure leaking 5x faster than core LLM providers, according to the State of Secrets Sprawl 2026.
A question worth separating out:
Q: Should security teams prioritise data unification or AI-driven response first in the SOC?
A: Data unification comes first. AI-driven response depends on a complete, normalised view of the environment, and without that foundation the platform can only automate fragments of the problem. Teams should get telemetry, correlation, and evidence quality right before expanding delegation.
👉 Read our full editorial: AI SOC agents still need unified data before autonomy