AI for the Physical World

This article originally appeared in the February 2025 issue of Security Business magazine. Feel free to share, and please don’t forget to mention Security Business magazine on LinkedIn and @SecBusinessMag on Twitter.

Artificial intelligence startups aren’t an uncommon occurrence on the CES show floor, so prowling the pre-CES Pepcom show, it was not a surprise to find Archetype AI. The surprise was the way the company is pioneering a new form of artificial intelligence – one that could prove to be security’s holy grail.

“We are building a large AI model that can understand the physical world through sensor data,” explains Archetype AI COO Brandon Barbello. “Everyone else is focused on data from the web, like images, texts, and audio, but there is a whole universe of sensor data from the real world. We are building an AI that can understand this sensor data, and thus understand the physical world and be useful for solving real-life problems.”

Archetype – a 9-month-old Palo Alto, Calif-based startup – calls the AI model Newton, and it leverages data from sensors that security pros are quite familiar with: Radar, cameras, microphones, thermometers, and other environmental sensors. It combines this data with a “Large Behavior Model” to unlock insights about the physical world in real time.

“Our goal is to work with existing sensors of any type,” Barbello explains. “We’ve developed a way to encode and map this sensor data into one AI model, embedding space with natural language. That makes it possible to talk to the sensors.”

While Barbello demoed Newton in a smart home setting – integrating it with a voice assistant to control lighting and temperature automatically based on a person’s activity – it isn’t difficult to see Newton make the leap from consumer to enterprise security.

Watch the Live Demo from CES

“Security, of course, is a great vertical for this, because there are lots of difficult behaviors and to understand intent and make predictions,” Barbello says. “We call [Newton] a ‘Large Behavior Model’ (instead of a large language model) because it is about giving the AI the context of all these different sensors.

“In a security context, these could be cameras, or microphones, or radar to understand what people are doing in a physical space and who has an intent that you might care about from a security point of view,” Barbello adds.

Indeed, Newton appears to promise the holy grail for the security industry: To be able to harness AI to analyze hundreds – if not thousands or millions – of sensors to truly be able to predict a potential security threat and stop it before something happens.

“What Newton makes possible is you can just take all those different sensor feeds, plug them with one API into one model, and have that one API pass you back the insights from the sensor fusion,” Barbello explains. “We can consolidate and simplify what it is for security to get value from their sensor systems for their customers.”

He adds: “We think that’s where this is headed. We call it a large behavior model because, unlike other kinds of AI that are just trying to classify things, we are looking at patterns over time to extrapolate what the next pattern will be, but in context. A knife in a kitchen and a knife in an airport mean two very different things – it’s about how you take that behavioral context and flag anomalies to normal behavior patterns.”