The Role of AI in Addressing
Public Safety Staffing Shortages

Public-safety agencies across the country continue to grapple with severe and persistent staffing shortages. ECCs remain particularly vulnerable, as vacancies among telecommunicators increase call-processing times, drive burnout, and degrade service levels. AI presents some of the most promising opportunities to relieve this pressure — although progress has been slower than many anticipated.

• AI technology is being used in some ECCs to intercept, classify, and triage nonemergency calls before they reach a human telecommunicator. This can significantly reduce call volume and free personnel to focus on higher-priority incidents.
• Transcription and real-time language translation capabilities have advanced substantially in the last year and are being leveraged to lessen the cognitive burden on telecommunicators, reduce call-processing times and improve service delivery, and eliminate errors.

An equally important opportunity lies in AI’s ability to boost human performance in high-stress, time-critical environments. AI-supported performance tools — particularly in the areas of quality assurance, training, and call analysis — already are making measurable strides, even if adoption remains slower than anticipated.

• AI can be used to evaluate every call shortly after completion, providing immediate feedback to telecommunicators and enabling training that is better aligned with real deficiencies.
• Though still at a nascent stage, AI can be used to perform telecommunicator voice analysis (tone, cadence, inflection) to flag potential burnout, cognitive overload, and indicators of an impending meltdown.

A year ago, agencies tended to view AI as a standalone technology —something they would “buy” as a dedicated tool, like transcription software or a separate analytics engine. Today, however, AI is no longer purchased as a product. It has become an embedded capability woven into the systems and software ECCs already rely on every day. This shift is transformational.

• Examples include AI-powered search and query features inside CAD systems; embedded transcription within call-handling platforms; voice-activated commands for navigating databases; summaries automatically generated from call or incident data; and predictive prompts that help dispatchers or call-takers surface relevant information faster.
  
• Historically, ECCs avoided deploying standalone AI tools because they required budget authorization and workflow redesign, i.e., telecommunicators don’t want to deal with yet another attention-sapping screen or application. In contrast, embedded AI arrives pre-integrated, with most of the technical work done by the system vendor.

AI is rapidly reshaping how public-safety agencies, and justice organizations process information and make decisions, to enhance speed, accuracy, safety, and situational awareness. At the same time, agencies are discovering that advanced automation and machine learning does not replace human judgment — rather, it amplifies it, enabling personnel to make faster, more informed decisions in high-stakes environments.

• AI’s ability to detect patterns across billions of data points enables automated identification of emerging threats or complex situations more quickly than humans can.
• AI’s multimodal capabilities — integrating text, audio, images, and video — enable robust decision support for ECC personnel and field responders, e.g., interpreting live surveillance footage or drone feeds to identify hazards or analyzing “alert-rich” environments (alarms, sensors, body-worn cameras) for decision-making cues.

Reliability is one of the most significant obstacles preventing public-safety agencies from fully embracing AI. These issues fall into several interconnected categories: accuracy, hallucinations, bias, reasoning limitations, prediction-driven behavior, and inconsistent performance across tasks. Each of these reliability shortcomings introduces risks that are unacceptable in environments where consequences can include delayed response, misdirected resources, legal exposure, or public mistrust.

• Up to 25 percent of AI responses may be incorrect, misleading, or entirely fabricated, depending on the use case. This is highly problematic because a single incorrect output could influence dispatching decisions, prioritization, or interpretations of critical information. Data bias, though improving, is a significant factor.

• AI models predict, but don’t “think.” They can replicate patterns but cannot replicate human reasoning; they struggle with multi-step logical deductions, and they may fail basic relationship or context inferences that a human would consider trivial.

While technical issues such as hallucinations and data quality pose quantifiable risks, human concerns — ranging from fear of job loss to skepticism about fairness — are far more difficult to overcome. They slow adoption, limit pilot success, and shape agency attitudes toward AI integration.

• Many public-safety professionals — especially veteran staff members — do not understand how AI works, what its limitations are, or how it produces results.
• Public-safety professionals are trained to avoid mistakes because errors can cost lives or lead to legal consequences. This makes them inherently skeptical of AI.
• Another factor that fuels skepticism concerns the current AI financial model, which is unsustainable — what happens when a 911 center depends on an AI tool and its vendor goes bankrupt?

While AI adoption is accelerating across the public-safety sector, the development of standards, policies, and operational guardrails has not kept pace. Agencies are experimenting with AI tools — sometimes intentionally, sometimes because AI has been silently embedded into the software they already use — but most do so without fully developed guidelines or governance structures. This gap increases organizational risk, slows adoption, and undermines trust.

• Unlike long-standing areas such as computer-aided dispatch (CAD) and call-handling, there are no consensus standards for using AI in the public-safety environment.
• AI requires human oversight and likely always will, but no framework exists for this purpose. Lacking defined oversight best practices increases liability risk and erodes staff confidence. Similarly, transparency requirements are not defined across the sector, e.g., agencies lack standards for explaining how AI reached conclusions.