Ambient.ai's CTO to security directors: if the AI made more alerts, it isn't assessing anything
In a Security Info Watch Q&A, Vikesh Khanna describes what the company means by agentic physical security, puts numbers on a ServiceNow deployment, and draws a line between automating alarm triage and automating a door lock.
Steve Lasky, editorial director of Security Info Watch, published a long Q&A with Ambient.ai co-founder and chief technology officer Vikesh Khanna on September 16, 2026, under the headline "Security Will Be Agentic." The title is Ambient's marketing line. The interview is more careful than the title.
Lasky's second question is the one security directors keep asking: every vendor claims to be AI-powered, so how do you tell what is real? Khanna's test is alert volume. If deploying AI produces more alerts than before, the system is notifying rather than assessing, and the operator's day has gotten worse. He says that pattern is common.
Detection versus reasoning
The distinction Khanna keeps returning to is between recognizing an object and understanding a situation. "Detection tells you a person is there. Reasoning tells you they have circled a loading dock three times at two in the morning," he said. Ambient's stack for doing that now includes Pulsar, an edge-optimized vision-language model launched in late 2025; an access-control product released in the spring to cut false door alarms; and, from August, Agentic Video Walls, which pick out the single most relevant event across every camera every 60 seconds, alongside a case management tool with semantic search. The company also says it doubled the number of cameras its edge appliance can handle.
The customer figures are the part of the interview most likely to be quoted back at competitors. At ServiceNow, Khanna says, the deployment produced a 94 percent reduction in false alarms and removed a little over 15,000 hours of triage. At SentinelOne, forensic investigations that took about an hour now take minutes. Ambient names Cisco and TikTok as customers, says new annual recurring revenue doubled and its customer count tripled since the Ambient Intelligence launch, and puts net revenue retention above 140 percent. Lasky does not press on any of these numbers.
Where the person stays
The most useful exchange is about limits. "Agentic does not mean autonomous," Khanna said, adding that whether to automate something, and how far, has a different answer for filtering alarm noise than for locking a door. He also argues that someone inside the customer's organization has to own AI decisions the way someone owns IT, because a vendor should not be the one deciding what a customer automates.
Khanna expects drones, ground robots and eventually humanoids to show up in security operations, with the agent supplying context to the machine and a human still making the call. The near-term claim, that a reasoning layer can turn a flood of detections into a handful of situations, is the one the ServiceNow numbers are meant to support.
Read the original at Security Info Watch.