Synthetic voices complicate verification for 911 dispatchers

Dispatchers at 911 centers now contend with a new layer of deception. Artificial intelligence tools allow prank callers to generate voices that mimic human hesitation, making it difficult for operators to identify fake bomb threats during time-sensitive interactions. This shift forces emergency workers to rely on more granular questioning techniques to validate incoming calls.

The issue gained visibility in August when Wheatland Union High School in Yuba County, California, received an overnight bomb threat. Police responded to the site, and classes were canceled for the day. No explosive was found, and no injuries were reported. Local media outlets reported that the caller used an AI-generated voice to demand $15,000 in a container box. Authorities identified the incident as part of a pattern of hoax calls, known as swatting, targeting schools across the country during the summer months.

Swatting involves calling 911 to report a false emergency, such as a bomb threat or active shooter, with the intent of drawing a large police response to a specific location. The tactic creates chaos and has resulted in fatalities and injuries in past incidents. A joint report by the Association of Public Safety Communication Officials (APCO) and the Public Safety Threat Alliance, published in June, analyzed data from more than 760 responses. The findings indicated that approximately 14% of dispatchers reported being targeted by calls using AI tools.

Mel Maier, CEO of APCO, noted that synthetic voices have become increasingly difficult to distinguish from real ones. He stated that earlier AI voices were easy to recognize, but current versions sound natural and include filler words like 'um' and 'uh.' Brandon May, an assistant professor of forensic psychology at the Florida Institute of Technology, added that the technology is inexpensive and accessible. He noted that he could clone his own voice for less than $20 using software from the company ElevenLabs. May observed that detecting these fakes is challenging even for experts, and he described the task as nearly impossible during a high-stakes call lasting 60 to 90 seconds.

In response, APCO is implementing training programs for dispatchers. The new protocols focus on criteria-based caller interrogations. This method involves asking specific questions that require detailed knowledge, making it harder for AI systems to pull answers from public databases or guess correctly. Maier emphasized that dispatchers will continue to treat every call as a genuine threat. Even if an operator suspects a call is a hoax, they will send resources to the location to ensure safety.

The availability of AI tools that can also simulate background sounds like gunfire or screaming adds to the complexity. May explained that these tools allow users to alter intonation and pitch, further blurring the line between reality and simulation. The training aims to equip dispatchers with the skills to navigate these technical challenges while maintaining rapid response times.