Splunk announces new AI agent governance and security tools

At the Splunk . conf26 conference in Denver, Cisco-owned Splunk announced several advancements aimed at managing the transition from AI assistants to autonomous AI agents. Key themes included the need for trust, cost control, and governance in enterprise environments.

Splunk introduced new capabilities for an ‘Agentic SOC’, designed to automate security detection, investigation, and response under human-defined policies. To address the risks of non-deterministic AI behavior, such as hallucinations and unexpected token consumption, the company unveiled ‘Splunk Agent Observability’. This solution utilizes small language models (SLMs) to provide real-time guardrails and monitoring across applications and infrastructure.

Additionally, Splunk introduced ‘Tokenomics’ to help organizations track and predict AI-related costs by visualizing token consumption and productivity metrics. The company also emphasized a decentralized data approach, using ‘Federated Search’ to analyze data where it resides—including AWS CloudWatch Lake and Databricks—rather than moving sensitive information to centralized AI models.