Feds Prep AI Cyber Drill, Ban DeepSeek Tools

The Federal Chief Information Security Officer's Council plans to bring agencies together in early November for a governmentwide tabletop exercise on AI-enabled cyberattacks. The drill will test whether the federal government can respond when an attack arrives. Cyberattacks using artificial intelligence increased by 89 percent in 2025 compared with 2024, according to the report cited by the council, and AI use in cyberattacks is increasing again this year.

The Exercise and the Threat

Mike Duffy, the acting federal CISO, said the drill will help agencies figure out what they actually have on hand. The question, he said, is whether agencies have the capabilities, processes and personnel needed to respond to cyber events. Speaking at ACT-IAC's Cybersecurity Summit on Oct. 14, Duffy described the planned exercise as the council's second all-government CISO-level tabletop. The first one, held the previous year, focused more broadly on readiness and resilience. This time, the scenario is narrower and more pointed.

"We are preparing for the second all full government CISO level tabletop exercise," Duffy said. "We had the first one last year to exercise what it means for not only individual agencies to exercise their incident response playbook, but what does it mean for all agencies?" He said council members would meet with security operations center directors to make sure the federal enterprise has what it needs.

Attackers now move faster than they used to. Duffy pointed to CrowdStrike's 2026 report, which found that hackers using AI-accelerated phishing and automated reconnaissance were shortening the time between initial access and damage. According to the report, AI had elevated less sophisticated threat actors and amplified the most advanced ones. Two problems are landing on the same government desks at once.

Policy Moves in Motion

Duffy said the Trump administration released a national cyber strategy in March. The Office of Management and Budget has concentrated on policies, procedures and capabilities for defending against attacks, including changes to cyber logging architecture policy. The Cybersecurity and Infrastructure Security Agency has also changed how it handles its known exploited vulnerabilities catalog, ranking flaws by factors such as likely exploitation and potential mission impact rather than treating every entry equally.

Agencies are applying those changes while working under a binding operational directive to secure and modernize endpoint devices. Duffy said a sprint within the CISO Council is testing how far agencies can get with existing resources before the fiscal year ends, including removing outdated edge devices and infrastructure. The long-term goal, he said, is to keep resilience going by repeatedly checking which systems need attention. Government has done plenty of one-time cleanups before. Now, the aim is to make the process recur.

The council has also brought back its AI working group, which focuses on security governance, and started a post-quantum cryptography working group in the previous month. Duffy described the post-quantum effort as a multiyear project requiring expertise and workforce migration. He named the National Institute of Standards and Technology and the Cybersecurity and Infrastructure Security Agency among the bodies working on what comes next.

Keep the Mission Running

Travis Rosiek, public sector chief technology officer at Rubrik, argued in a Federal News Network commentary that agencies should plan to keep critical missions running during an attack, not just try to stop every intrusion. Prevention alone won't carry the load. Rosiek wrote that critical infrastructure, operational technology and networked communications create a vulnerable interface between U.S. defense and civilian sectors.

He cited the Office of the Director of National Intelligence's 2026 Annual Threat Assessment, which said adversaries were exploiting supply chains to compromise software source code and pre-position themselves inside federal networks. As Rosiek described it, the goal is persistent access that could shut down critical services, disable power grids and disrupt military readiness. He also pointed to the self-replicating Shai-Hulud worm, which hackers deployed into developer pipelines in 2025. It got past perimeter defenses, harvested cloud credentials and spread into private and public networks.

Rosiek's central complaint is that disaster recovery plans built for floods, power outages and hardware failures don't handle a deliberate cyberattack. He urged agencies to design systems that can isolate breached segments while keeping core functions alive, test recovery plans with security, cloud, IT and backup administrators, and inventory their data, identify owners and map dependencies. Rigorous testing, defined metrics and executive ownership, he wrote, could make recovery repeatable rather than improvised.

DeepSeek Ordered Out

The Department of Homeland Security has ordered federal civilian agencies to remove artificial intelligence developed by China's DeepSeek from their systems and barred future purchases of its products and services. The order, issued in a DHS notice, also covers third-party services built on DeepSeek. The notice said the company's models "present multiple threat vectors" and that use of its products "may result in the transfer of sensitive data to the [People's Republic of China] if used by US government employees or vendors."

DHS acted under the Federal Acquisition Supply Chain Security Act of 2018, after reviewing a recommendation and supporting evidence from the Federal Acquisition Security Council. The notice didn't detail the specific security concerns, and it set no deadline for agencies to comply. Program managers will have to guess how quickly removal must happen.

Liang Wenfeng founded DeepSeek, whose models offer cheaper alternatives to products from Anthropic and OpenAI. The article notes that performance had lagged leading U.S. models, though the gap had narrowed over the previous year by some measures. DeepSeek is an open-weight model, meaning users can customize it more easily than the closed models from Anthropic, Google and OpenAI. Its arrival early the previous year with cheaper models contributed to a $1 trillion rout in major U.S. technology stocks, including Nvidia, Microsoft and Tesla. The Commerce Department had already banned DeepSeek from government devices the previous year, the U.S. Navy told its members not to use the model over potential security concerns, and New York state barred it from government devices and networks, citing "serious concerns" about data privacy and censorship.

Data Strategy Gap

In a separate Federal News Network commentary, Taka Ariga, vice president for data policy and impact at the Data Foundation, called for a national data strategy. Ariga wrote that the United Kingdom published a National Data Strategy in 2020 and that China built national policy and a dedicated agency around data as a strategic asset, while the United States has no national strategy. Data.gov launched in 2009 and provides public access to many federal, state and local open datasets. The Foundations for Evidence-Based Policymaking Act, signed by President Donald Trump in 2019, requires a standard application process through which researchers can request restricted federal data. The Federal Data Strategy, released that same year, encouraged agencies to manage data better and drew on input from industry and researchers.

Ariga argued that generative AI has made public data more valuable and more sensitive. States and localities are building data capacity, private and philanthropic money is flowing into projects that combine public and private data, and expectations about privacy are shifting as information gets combined across sectors. The Data Foundation has launched a community-led effort to guide a national data strategy, planning a living document open to feedback and critique. Ariga said the effort isn't meant to produce a new federal strategy. It is meant to build on the existing federal foundation and extend it across society. The discussion draft and updates will be posted at NationalDataStrategy.us.

Why This Matters:

The federal cyber posture now depends on several moving parts: a drill that hasn't happened yet, directives with no hard deadline for removing a foreign AI tool, and recovery plans that many agencies weren't built to run during a deliberate attack. Attacks using AI rose 89 percent in 2025 by the report's measure, and the government is assembling its response while that pace continues. Fiscal pressure is part of the picture too, since Duffy's sprint is testing what agencies can accomplish with existing resources before the fiscal year closes. The DeepSeek order raises a supply-chain question touching procurement budgets and private vendors building on the same models. Whether these efforts become routine, repeatable habits rather than one-off responses will determine whether the government can keep critical services running when the next intrusion arrives.