Investors bet on agent identity as the next big security market

Cybersecurity investors say enterprises are plugging AI agents into business-critical systems without the identity, access controls or audit trails needed to govern them, according to interviews published by The Register on 19 September 2026. Two investors — Matt Hartman, chief strategy officer at Merlin Group, and Todd Graham, managing partner at Microsoft's M12 venture fund — described a market in which security is lagging well behind deployment.

The Register reported that the gap is visible in a rise in incidents involving agents hacking organisations and people, and other mishaps with agents gone rogue. "On one hand, we shouldn't be surprised that increasingly capable agents are finding creative and sometimes unexpected ways to accomplish their objectives," Hartman told the publication. "On the other, we can't accept harmful behaviour as inevitable or unmanageable."

Graham framed the problem as a repeat of earlier technology cycles. "Every time we've built a new piece of infrastructure, we've conveniently forgotten the security," he said. He pointed to the commercial results of that pattern: "If laptops were default secure, we wouldn't have CrowdStrike. If the cloud was default secure, we wouldn't have Wiz. If identity wasn't default secure, we wouldn't have a bunch of Active Directory add-ons and Okta."

What differs this time, according to Graham, is pace. Cloud adoption took years; organisations are moving much faster to give agents access to sensitive data and applications, while lacking a reliable way to manage, secure or even identify the agents already operating in their systems. "This is a transition happening month over month," he said, describing recent incidents as "a wake up call".

Hartman, who joined Merlin after senior roles including at the US Cybersecurity and Infrastructure Security Agency, said his firm is focused on the control layer around agent behaviour. That means identity for non-human actors, explicit limits on what an agent can access and do, and an audit trail of actions taken on an organisation's behalf. "Agencies aren't just asking how to adopt agents, they're asking how to constrain them and prove what one did at 2 AM on a Tuesday," he said.

Both investors warned that building a product is no longer the hard part. "AI has made it faster and cheaper than ever to build a product, so the bar for differentiation keeps rising," Hartman said. Graham said many founders are "thinking way too small" and solving "a sliver of the problem". A Fortune 500 chief information security officer, he argued, will not buy 15 separate tools; agent identity products will need to cover governance, access control and authorisation in the way human identity stacks already do.

The Register noted that non-human identity is already a sore point. Service accounts are a prime target for attackers because they typically carry high privileges and passwords that never expire, a problem that predates agents.

Graham also flagged AI endpoint security — effectively antivirus or endpoint detection and response for AI — as an underserved area, suggesting executives would face questions if a breach exposed the absence of such controls. He expects incumbent endpoint vendors to move into the space, but said the field remains early enough for newcomers.

Asked about recent agent misbehaviour, Graham said he was "worried", adding: "If left unencumbered, if left to its own devices, I am very concerned about where the endpoint is for this."