Mortada Ayad, VP- META, Delinea, told GEC Newswire that machine identities already outnumber people 46 to 1 in some organisations, and that the old habit of controlling access has stopped working.
Every security conference in the region has had its tagline, and Mortada Ayad, VP- META, Delinea, has watched them rotate. Zero trust, then identity, now AI security. He said the underlying problem has not moved. “You always start with a compromised identity that has more or less permissions, and then from there, we escalate, and we’re able to move laterally,” he said.
An AI agent frequently operates under the same credential as the person who deployed it, which leaves security teams with no way of telling who acted. “If we have the same session, myself or my agent doing something on my behalf, we need to be able to differentiate the two,” Ayad said.
The ratio nobody in the region has caught up with is 46 to 1
Delinea research at the start of this year put machine and AI identities at roughly 46 for every human, a figure Ayad said had risen since. Delinea did not release the study’s sample size or methodology. He resisted the idea that regional organisations are lagging, pointing to a governance gap that widens with the ratio.
Ayad set out 3 controls organisations should complete before buying anything. Discover every AI identity already running, which he said existing identity providers can do without new licences. Attach each one to a named human, so ownership is never in question. Then grant only the permissions the task requires.
His chosen analogy is deliberately unglamorous. “Treat those AI agent as interns that will join the organisation,” he said. Interns arrive mapped to a department, hired for a defined task, and nobody hands them every key.
You cannot tell an autonomous agent what to do, so stop trying
Accountability has to sit with a person. “If something goes wrong, there must be a human made accountable for that action,” Ayad said, adding that the vendor’s job is to keep the blast radius small by ensuring the credential in use carries minimal privilege.
He distinguished delegated agents, which borrow a human credential, from autonomous ones that run to a goal without supervision. Neither obeys a script. “You give the agent a goal, and it will do everything it believe is necessary to reach that goal,” he said.
One lever remains. “You have to control at the action,” Ayad said, describing a shift away from granting standing privilege at the point of access towards releasing permission only at the moment it is used.
