Autonomy Before Control: The 2026 C-Suite Report on AI Agent Risk

We surveyed 500 C-suite leaders in the US and UK. The finding is clear: AI agents are going live faster than the controls meant to govern them.

AI agents are no longer just support tools. They are making decisions and taking action across customer data, code, and financial systems.

Our new research shows leaders are ready to hand over real authority, even as incidents pile up and security gaps force projects to slow down or stop.

Key findings:

  • 92% would allow AI agents to take at least some independent action
  • 30% of leaders who would let AI make decisions would allow agents to approve financial transactions, modify code, or access customer data without prior human approval
  • 77% of organizations using or exploring AI have experienced an incident where a system or agent acted outside its intended or authorized behavior
  • 36% of US leaders would let agents approve financial transactions on their own, compared with 24% in the UK

Visibility is the starting point, not the finish line. Knowing an agent exists does not tell you what it can reach, what it is allowed to do, or how it will behave.

The Salt Agentic Security Platform discovers your agents, MCP servers, and APIs, governs their connections and permissions, and monitors behavior at runtime, so you can scale AI with confidence instead of slowing it down.

Download the Full Report