This post is also available in: heעברית (Hebrew)

As mass shootings and gun violence continue to climb in the US, security has become one of the highest priorities for many organizations. A variety of new technologies have been developed to help prevent, detect and respond to these threats — including gunshot detection and improved security cameras.

An original approach has been taken by AI security camera startup Athena Security. The company uses computer vision and other technologies to monitor and detect potential threats, such as someone carrying a weapon. The system automatically sends an alert to authorities to expedite response. It also masks the people it captures on video to avoid profiling based on race and other factors before letting AI analyze the footage.

The company told that it has nearly 1,000 cameras out in the field with thousands more being deployed at schools, businesses small and large, and places of worship.

COO Chris Ciabarra tells about the new technology: “We don’t report sounds, we use AI cameras to report instantly the first moment a gun is drawn or crime is committed, that can make Athena Security a preventative tool or even a life saving tool because police can be on their way a significant amount of time earlier than once a gun is fired or a crime escalates and is reported by someone,” Ciabarra said.

It’s all about training. The company has been training its AI with actors who play out different scenarios that could lead to violence. The system, in turn, learns to identify real threats and avoid false alarms. “The thing holding A.I. back was that a false positive means you can’t be deployed, you can’t have a system crying wolf,” Ciabarra said. “So we thought about how to best/better train the A.I. brain, and, instead of having it watch action movies, we realized we needed to have trained actors making the moves and wielding the weapons in order to get up to the 99% accuracy level. After that, we had police and SWAT teams put the A.I. brain through its paces and now we’re getting on-the-job training because every frame of every camera every minute is helping train the system to recognize everything more accurately.”