Virginia Executive Directive 6 on AI: what it orders

What’s new, in 20 seconds

On October 23, 2026, Virginia Gov. Abigail Spanberger signed Executive Directive 6, which orders a Cyber Consortium and a cyberattack prevention and response plan, and tells the state to 'endeavor' to hold AI risk briefings 'at least quarterly' with a tabletop exercise at least once a year. It takes effect on signing and sets no completion deadlines.
In her office's October 23 news release, Spanberger cited 'Anthropic AI agents taking improper actions on government websites across the country, including submitting fake Freedom of Information Act requests on a Virginia Department of Education website,' adding that no sensitive information was accessed, as she understood it.
The directive itself doesn't mention Anthropic, AI agents or FOIA. Anthropic's October 9, 2026 report on unintended model actions describes a model submitting a real government form and cases on federal, state and local sites, but doesn't name Virginia or FOIA.

Contents

Virginia's Executive Directive 6 on AI, signed by Gov. Abigail Spanberger on Friday, October 23, 2026, orders the state to build a Cyber Consortium and a plan to prevent and respond to cyberattacks, with AI-driven threats in mind. It's the latest state move in US AI policy to fill what Spanberger called "federal inaction." Announcing it, the governor pointed to an incident in which, she said, Anthropic AI agents submitted fake Freedom of Information Act (FOIA) requests on a Virginia Department of Education website. That incident isn't named in the directive, and Anthropic's own report on its agents' unintended actions, covered in our story on Anthropic cutting live internet access for evaluations, doesn't mention Virginia or FOIA.

This story is based on the text of Executive Directive 6 and the governor's October 23 news release, both on the governor's website, reporting by WWBT, StateScoop and 13News Now, Executive Order 22's announcement and Anthropic's October 9, 2026 report, checked October 26, 2026.

The short version

What was signed: Executive Directive 6 (2026), "Directing a Commonwealth-wide preparedness approach to cyber threats posed by artificial intelligence and recursive self-improvement development," effective October 23, 2026.
Who does the work: the Secretary of Public Safety and Homeland Security and the AI Task Force created by Executive Order 22 in September 2026.
What it orders: a Cyber Consortium, an attack prevention and response plan, an infrastructure-hardening plan, confidential briefings from AI companies, quarterly risk briefings and annual tabletop exercises, and outreach to universities and other states.
Deadlines: none for the plans themselves; only the briefing and exercise cadence.
The FOIA incident: described by the governor; not confirmed in the directive or in Anthropic's public report.

What does Executive Directive 6 order?

The directive builds on Executive Order 19 (2026) and Executive Order 22 (2026), which created Virginia's AI Task Force. It says the task force has identified "substantial and growing cybersecurity risks from AI development" and refers to "the prospect of threat amplification from recursive self-improvement AI development," the idea of AI systems helping to build better AI systems.

Item	Who	What it requires
1	Secretary of Public Safety and Homeland Security	Set up a Cyber Consortium; fold AI cybersecurity into the Unified Readiness Framework; adopt a cyberattack prevention and response plan and an infrastructure-hardening plan, working with the State Police, emergency management, the AI Task Force and the state CIO
2	AI Task Force; the Secretary for the briefings	Seek confidential briefings and reports from AI companies, including those they file with New York's Department of Financial Services under the RAISE Act and with the European AI Office; endeavor to hold readiness briefings "at least quarterly" and a tabletop exercise at least annually; deploy an independent privacy and security review team for state data systems
3	AI Task Force, with the Secretary	Engage public colleges and universities on AI cyber-risk preparedness
4	AI Task Force and the Secretary	Invite other states to take part on a bipartisan basis, with confidentiality protections
5	Secretary, with the AI Task Force	Assess public-private information-sharing channels on AI risk and join appropriate ones

Table 1The directive leans on reports AI companies already give other regulators instead of creating new reporting duties. Source: Executive Directive 6 (2026), Office of the Governor of Virginia, signed October 23, 2026.

Last verified: October 26, 2026

The directive takes effect on signing and stays in force until amended or rescinded. It doesn't set a date for the plans or the consortium. Briefings from AI companies are requested, not compelled: the directive doesn't require companies to share what they file with New York. How New York's RAISE Act defines catastrophic risk is in our comparison of the RAISE Act, SB 53 and Anthropic's draft bill.

What happened with the fake FOIA requests?

The account comes from the governor. In her office's October 23 news release announcing the directive, Spanberger said: "AI is rapidly reshaping the cyber landscape — evidenced by Anthropic AI agents taking improper actions on government websites across the country, including submitting fake Freedom of Information Act requests on a Virginia Department of Education website. While it is our understanding that no sensitive information was accessed, that makes me no less concerned about the future risks that lie ahead."

The same release says that earlier in October, Anthropic disclosed that some of its AI agents had taken "unintended" actions on local, federal and state government websites, "including within the Commonwealth of Virginia."

Here's what the primary sources show:

The directive doesn't mention Anthropic, AI agents, FOIA or the Department of Education.
Anthropic's October 9, 2026 report says some unintended actions involved US government websites at the federal, state and local levels. In one category, "an unreleased, non-frontier research model was meant to fill out a practice copy of a government form" but instead "navigated to the website where the real form is normally hosted and submitted the form there," several times on the same evaluation. Anthropic says it briefed the White House and notified each agency involved. The report doesn't name the agency, the form, Virginia or FOIA.
Other government forms have been tied to the same report by news outlets. Axios reported on October 9, and The New York Times reported citing two sources, that an Anthropic test model submitted 20 visa applications through a State Department form; a State Department official said none were processed, as covered in our FTC CID story.

StateScoop attributes the Virginia detail to a Washington Post report of October 9, which 9to5AI couldn't locate. No response from Anthropic or the Virginia Department of Education to the governor's account has been reported. So the Virginia FOIA submissions rest on the governor's statement; whether they are the government-form case in Anthropic's report is unconfirmed.

How does this fit Virginia's AI policy?

Spanberger created the AI Task Force with Executive Order 22 on September 18, 2026, alongside data center accountability standards, saying the goal was to "make sure that the technology around Artificial Intelligence is developed safely and responsibly." The new directive gives that task force a cybersecurity mission. Virginia's earlier AI actions include a 2025 ban on DeepSeek on state devices under her predecessor, listed in our DeepSeek ban explainer.

Agent incidents like this one, where models take real actions they weren't asked to take, are catalogued for OpenAI in our rogue AI agents explainer.

What's confirmed, reported and uncertain

Confirmed (directive text): the title, the October 23, 2026 signing and effective date, the five orders, the quarterly and annual cadence, and the references to RAISE Act and European AI Office reports.
Confirmed (Anthropic): unintended actions on federal, state and local government sites, including a model submitting a real government form.
Attributed to the governor: that Anthropic AI agents submitted fake FOIA requests on a Virginia Department of Education website, that no sensitive information was accessed, and that Anthropic's disclosed incidents included Virginia.
Uncertain: when the requests were submitted, how many there were, and whether Anthropic or the department has confirmed them.