Blab AI
U.S. States Urged to Bolster Cyber Defenses Amid Growing Threat of AI-Powered Attacks
With artificial intelligence advancing at breakneck speed, state and local governments are racing to fortify their cyber defenses.
Officials warn that cutting‑edge AI models could give attackers the tools to mount sophisticated, zero‑day attacks on a massive scale.
A recent NASCIO‑Deloitte survey of state chief information security officers (CISOs) revealed that just 22 % feel confident in protecting public data—a dramatic drop from 48 % in 2022.
The study, carried out by NASCIO with Deloitte, also highlighted that state cybersecurity budgets are flat or shrinking, while dependence on third‑party vendors continues to rise.
Utah’s chief information officer Alan Fuller told Pluribus News, "The world is dangerous now, and it’s about to get way more dangerous as these frontier models become more widely available." He added that Utah’s networks are scanned over a billion times daily, yet its AI‑driven monitoring may fall short against attacks that could materialize within six to nine months.
The danger became starkly apparent in September, when a suspected Chinese state‑sponsored group reportedly deployed Anthropic’s Claude Code tool against technology firms, banks, chemical manufacturers, and governments. Anthropic called the incident "the first documented case of a large‑scale cyberattack executed without substantial human intervention" and warned that "the barriers to performing sophisticated cyberattacks have dropped substantially." The report added that agentic AI systems can replicate the work of entire hacker teams—from system analysis to exploit code generation.
In April, Anthropic revealed that its new model, Claude Mythos, was too risky for public release due to its capacity to discover and exploit software vulnerabilities. The company has supplied Mythos to more than three dozen technology firms to help them uncover and remediate security gaps. A New York Times report noted that Mythos has already identified thousands of high‑severity vulnerabilities in major operating systems and web browsers.
Following Mythos’ findings, Anthropic launched Project Glasswing—a program aimed at "securing the world’s most critical software." The initiative now includes infrastructure operators in more than 15 countries and has identified over 10,000 serious security gaps to date.
In response to the escalating threat, President Donald Trump issued an executive order last week directing the Department of Homeland Security to "facilitate access to cybersecurity tools and services" for state and local governments and operators of critical infrastructure, including utilities and hospitals.
OpenAI has also joined the fray, announcing a limited preview of GPT‑5.5‑Cyber—a model designed for entities charged with safeguarding critical infrastructure. The preview aims to help defenders spot vulnerabilities before attackers do.
Despite these initiatives, state CISOs continue to express alarm. The NASCIO‑Deloitte survey placed AI‑enabled attacks in the top three concerns, alongside phishing and third‑party contractor breaches. Confidence that state information assets are shielded from threats originating within local governments and public higher‑education institutions fell from 59 % in 2022 to 26 %.
Nevertheless, 23 states have already deployed generative AI to bolster their cybersecurity operations, and 21 more intend to do so within the next year. Roughly one‑fifth of states are adopting a "whole‑of‑state" approach, offering umbrella support to entities beyond state government.
Funding remains a hurdle. State budgets are strained, and federal support for the State and Local Cybersecurity Grant Program is scheduled to lapse in January 2026. State leaders have urged Congress to reauthorize the $1 billion program.
Some states are also pursuing legal safeguards. California and New York have enacted laws that compel AI developers to address catastrophic harms, including large‑scale cybersecurity attacks. Illinois has passed a bill that is now headed to the governor.
In short, state and local governments confront a rapidly evolving threat landscape in which AI can accelerate the discovery and exploitation of software vulnerabilities. New AI tools and government initiatives seek to give defenders a head start, but confidence in public data protection remains low, and budget constraints could slow the pace and scale of needed upgrades.
Over the coming months, further collaboration between AI firms and public‑sector agencies, additional legislative proposals, and the continued rollout of AI‑driven security tools are expected. The success of these measures will decide whether states can keep pace with attackers who are increasingly able to harness frontier AI models.
Officials warn that cutting‑edge AI models could give attackers the tools to mount sophisticated, zero‑day attacks on a massive scale.
A recent NASCIO‑Deloitte survey of state chief information security officers (CISOs) revealed that just 22 % feel confident in protecting public data—a dramatic drop from 48 % in 2022.
The study, carried out by NASCIO with Deloitte, also highlighted that state cybersecurity budgets are flat or shrinking, while dependence on third‑party vendors continues to rise.
Utah’s chief information officer Alan Fuller told Pluribus News, "The world is dangerous now, and it’s about to get way more dangerous as these frontier models become more widely available." He added that Utah’s networks are scanned over a billion times daily, yet its AI‑driven monitoring may fall short against attacks that could materialize within six to nine months.
The danger became starkly apparent in September, when a suspected Chinese state‑sponsored group reportedly deployed Anthropic’s Claude Code tool against technology firms, banks, chemical manufacturers, and governments. Anthropic called the incident "the first documented case of a large‑scale cyberattack executed without substantial human intervention" and warned that "the barriers to performing sophisticated cyberattacks have dropped substantially." The report added that agentic AI systems can replicate the work of entire hacker teams—from system analysis to exploit code generation.
In April, Anthropic revealed that its new model, Claude Mythos, was too risky for public release due to its capacity to discover and exploit software vulnerabilities. The company has supplied Mythos to more than three dozen technology firms to help them uncover and remediate security gaps. A New York Times report noted that Mythos has already identified thousands of high‑severity vulnerabilities in major operating systems and web browsers.
Following Mythos’ findings, Anthropic launched Project Glasswing—a program aimed at "securing the world’s most critical software." The initiative now includes infrastructure operators in more than 15 countries and has identified over 10,000 serious security gaps to date.
In response to the escalating threat, President Donald Trump issued an executive order last week directing the Department of Homeland Security to "facilitate access to cybersecurity tools and services" for state and local governments and operators of critical infrastructure, including utilities and hospitals.
OpenAI has also joined the fray, announcing a limited preview of GPT‑5.5‑Cyber—a model designed for entities charged with safeguarding critical infrastructure. The preview aims to help defenders spot vulnerabilities before attackers do.
Despite these initiatives, state CISOs continue to express alarm. The NASCIO‑Deloitte survey placed AI‑enabled attacks in the top three concerns, alongside phishing and third‑party contractor breaches. Confidence that state information assets are shielded from threats originating within local governments and public higher‑education institutions fell from 59 % in 2022 to 26 %.
Nevertheless, 23 states have already deployed generative AI to bolster their cybersecurity operations, and 21 more intend to do so within the next year. Roughly one‑fifth of states are adopting a "whole‑of‑state" approach, offering umbrella support to entities beyond state government.
Funding remains a hurdle. State budgets are strained, and federal support for the State and Local Cybersecurity Grant Program is scheduled to lapse in January 2026. State leaders have urged Congress to reauthorize the $1 billion program.
Some states are also pursuing legal safeguards. California and New York have enacted laws that compel AI developers to address catastrophic harms, including large‑scale cybersecurity attacks. Illinois has passed a bill that is now headed to the governor.
In short, state and local governments confront a rapidly evolving threat landscape in which AI can accelerate the discovery and exploitation of software vulnerabilities. New AI tools and government initiatives seek to give defenders a head start, but confidence in public data protection remains low, and budget constraints could slow the pace and scale of needed upgrades.
Over the coming months, further collaboration between AI firms and public‑sector agencies, additional legislative proposals, and the continued rollout of AI‑driven security tools are expected. The success of these measures will decide whether states can keep pace with attackers who are increasingly able to harness frontier AI models.
