Companies are running AI agents but few can say who is responsible

Companies are putting AI agents to work faster than they are deciding who is responsible for them. A new survey from Gravitee, a company that sells software for managing AI agents, found that fewer than 1 in 10 organizations have a named person formally accountable for what their agents do. An AI agent is software that takes actions on its own, such as reading records or sending requests, instead of just answering questions.

What the survey found

Gravitee surveyed 900 senior technology leaders in the US and UK in October 2026. They work in financial services, healthcare, government, telecommunications, manufacturing, and travel and transport. Over half, 52%, said they had an agent-related security or privacy incident in the past year. Only 17% give each agent its own identity, which is like a separate login, so it is clear which agent did what.

Growth is outpacing control

The number of organizations running more than 150 agents rose from 7% to 16% since Gravitee's April 2026 survey. Confidence in handling incidents has fallen over the same six months, while the controls around agents look the same. The survey also found agents are working in chains.

• 97% run agents that call other agents.
• 40% connect to agents belonging to other organizations.
• Only 27% define who is accountable across that chain.

When something goes wrong

More than half of respondents, 58%, have had to cut off an agent's access during a live incident. Afterward, only 25% held a named person responsible. And 17% changed nothing. Gravitee's CEO, Rory Blundell, put it this way: most organizations cannot say which agent it was, what it was allowed to do, or who is responsible.

Why it matters to you

The survey names banks, hospitals and government departments as places where agents are acting. If your data sits with a business like that, an agent mishap could touch it. The survey does not say how many incidents involved customer data, or how serious they were. It is also a vendor's own research, and Gravitee sells products in this area, so the findings are best read with that in mind.

What the company suggests

Gravitee's technology chief, Linus Håkansson, says to manage agents like an employee with access to sensitive systems. Give each one an identity, limit what it can do, keep a record, and be able to revoke access within minutes. The company's framework lists six principles: Identity, Sponsor, Role, Authority, Oversight and Recourse. Together they answer what acted, on whose authority, within what limits, and how it can be stopped.

What to do

If you use AI agents in your own business, name one person who answers for each. Give each agent its own login, limit what it can reach, and know how to switch it off quickly. If a company holds your data, it is fair to ask who is accountable when one of its agents goes wrong.