Marietta staff outline expanded use of AI for network security, call center and customer service

Marietta staff presented a detailed update on projects that use artificial intelligence to support network security, customer call handling and other municipal services during the City Council agenda review session.

The update, given by a staff member identified in the meeting as Ronny, said the city already uses machine-learning tools at the network perimeter to watch “what's normal behavior” and flag anomalies. He said those systems initially sent alerts and — after training — have been allowed to take some autonomous actions overnight. “It'll run 24x7,” Ronny said, explaining the advantage over human monitoring limited to business hours.

Nut graf: City leaders said they expect those AI tools to reduce staff workload and shorten citizen wait times while also strengthening cybersecurity. Staff outlined a multi-pronged rollout that includes network defenses, call-center enhancements, supervisor-assist features and internal chatbots for employees.

Staff described the call-center project for the municipal water and light works (BLW) as a phased deployment. The call takers were moved to a cloud platform and staff plan to enable AI features by late 2025. Ronny said one of the first features will be call summarization: the AI will transcribe and summarize a call so a customer-care worker can review and sign off, saving “several minutes per call.” He added supervisors will be able to flag exemplary calls in the system to build a training library.

Planned features listed by staff include "whisper mode," which would allow a supervisor to coach a call taker during a live call without the caller hearing the supervisor, and new email and chat queues that AI will help triage. Staff also described a “chatbot for the call takers” so employees on a call can query the AI in-session for policy or billing guidance.

City staff said AI training will require time and data ingestion: the system must ingest past calls, policies and answer sets before it can reliably summarize calls or assist employees. Ronny framed the timeline as months of work and testing, with multiple features targeted for activation “by the end of the year” and broader adoption through the following months.

Councilmembers asked about risks and behavior. Ronny acknowledged widely publicized incidents in which public LLMs produced unsafe responses but said the city is working with closed models and has not observed aggressive or unexpected behavior internally: “We have not experienced that yet,” he said. He also flagged the need to keep data clean and to control what the AI ingests.

Staff said the same tools could be tested for dispatch and after-hours coverage and might reduce some outsourced after-hours call handling if experiments prove reliable. They also proposed applications for HR and permitting: ingesting benefits and pension data so employees and residents could query routine questions without a staff referral.

On security, Ronny said staff are exploring tools that simulate voice-based phishing so employees can be trained to recognize deepfake phone calls and other social-engineering attacks. “The bad actors are out there thinking exactly the same thing,” he said.

Ending: Staff committed to periodic updates to the council as features are enabled and to continue restricting sensitive product details to avoid exposing operational specifics in public minutes.