AI agents are being deployed rapidly despite weak security controls, leading to incidents in which they hack organisations, access sensitive systems or behave unpredictably. Investors argue this is both a serious risk and a major opportunity for startups to build security tools that govern agents before harmful behaviour becomes normalised.

The proposed solutions include non-human identity management, strict access limits, behavioural governance and audit trails showing what agents did and when. Unlike slower cloud adoption, organisations are introducing agents within months and granting them access to critical data, while investors warn that successful startups must offer broad, differentiated products rather than narrow tools that leave security teams managing multiple disconnected systems.