In February 2021, someone gained remote access to the control system at the Oldsmar water treatment plant and increased the amount of sodium hydroxide being added to the water. An employee saw the change and corrected it before the treatment process was affected.

The subsequent federal cybersecurity advisory identified weaknesses including poor password security and an outdated operating system. Investigators said desktop-sharing software may also have provided the access point.

Five years later, that incident reads differently.

OpenAI and more than 100 companies, including Microsoft, Google, Amazon Web Services, Cisco, CrowdStrike, Cloudflare and IBM, have signed an open letter calling for a global surge in cyber defense. They warn that increasingly capable AI models will make sophisticated cyberattacks faster and more widely accessible in the coming months. Hospitals, water systems, governments and other critical infrastructure are specifically identified as being at risk.

Oldsmar offers a useful local example because the vulnerabilities were fairly ordinary. Old software. Weak password practices. Remote access. These are the kinds of weaknesses buried throughout large organizations that have accumulated technology over decades.

AI can search for them at a scale humans cannot.

An attacker can use an AI agent to examine systems, look for misconfigurations, test possible entry points and pursue vulnerabilities continuously. The level of expertise and time required to conduct that work is falling.

Fortunately, those economics work for defenders too.

AI can help a municipal cybersecurity team inventory systems, review configurations, identify excessive permissions, examine code, triage alerts and prioritize vulnerabilities. OpenAI says it is already using AI to continuously probe its own infrastructure for potential attack paths and triage most initial security alerts before involving people. Its new cybersecurity tools are designed for tasks including vulnerability discovery, secure code review, malware analysis, incident response and patch validation.

That capability could be particularly valuable in local government because cybersecurity expertise is difficult to scale.

St. Petersburg operates water and wastewater systems, public safety infrastructure, financial systems and a growing collection of digital services. Those systems also connect to vendors, contractors and other government agencies. Each connection expands the environment that a relatively small group of people must protect.

AI gives those people additional capacity.

There is also a timing problem that deserves attention. A widespread wave of AI-enabled attacks would create simultaneous demand for cybersecurity professionals, incident-response firms and specialized vendors. Organizations that wait until an incident to assemble that expertise could find themselves competing with hundreds or thousands of other organizations trying to do exactly the same thing.

Cybersecurity has always required preparation. AI increases both the potential scale of an attack wave and the amount of defensive work organizations can complete beforehand.

That is the useful idea behind what OpenAI calls the “defenders’ window.” The technology capable of making attacks easier is already capable of helping security teams find weaknesses, prioritize them and accelerate remediation. The opportunity exists before those offensive capabilities become commonplace.

A Practical Model for AI Cyber Readiness

Inventory the attack surface.
Identify systems, software, privileged accounts, remote access, vendor connections and legacy technology across the organization.

Expand the existing team.
Give cybersecurity staff governed AI tools that can help inspect configurations, analyze alerts, identify vulnerabilities and prioritize work.

Fix exposure before volume arrives.
Address the highest-risk weaknesses while cybersecurity specialists and vendors still have capacity to help.

Practice the response.
Test incident procedures, AI-assisted investigations and recovery processes before an actual attack requires them.

The Oldsmar employee who noticed a chemical setting suddenly changing had time to intervene. The next generation of attacks may move much faster and may arrive in much greater numbers.

Local governments have a window to prepare their people and systems for that environment. They should use it.