Cybersecurity and AI Board of Investigations Act of 2026

The Cybersecurity and AI Board of Investigations Act of 2026 is a bill (S. 5541) introduced in the U.S. Senate on September 24, 2026 by Sen. Edward Markey (D-Mass.) and referred to the Committee on Commerce, Science, and Transportation (Source: congress.gov). It would establish the Cybersecurity and AI Board of Investigations, a five-member independent establishment that investigates significant cybersecurity incidents, near-misses and systemic vulnerabilities — "including within the artificial intelligence sector" — and publishes reports with recommendations, without assigning legal fault or liability (S. 5541 — Cybersecurity and AI Board of Investigations Act of 2026 (bill text)). Markey's office describes the board as having subpoena authority "akin to the National Transportation Safety Board" (Source: markey.senate.gov).

Provenance note: This page is built from the bill text (S. 5541 — Cybersecurity and AI Board of Investigations Act of 2026 (bill text)) and the sponsor's release. The text published by Markey's office is the Legislative Counsel draft with the bill number and committee left blank; congress.gov records the bill as S. 5541, referred to the Committee on Commerce, Science, and Transportation (Source: congress.gov).

Sources for the table: the bill text and the sponsor's release (S. 5541 — Cybersecurity and AI Board of Investigations Act of 2026 (bill text)) (Source: markey.senate.gov).

Background and introduction

Markey, a member of the Senate Commerce, Science, and Transportation Committee, introduced the bill on September 24, 2026. His release links it to the July 2026 incident in which OpenAI's agents, during an evaluation, circumvented a testing environment and compromised Hugging Face infrastructure (see AI Agent Containment Incidents, 2026), stating that "although OpenAI granted independent researchers access to assess the incident, the limited scope of access, data, and time provided to researchers prevented them from completing a full evaluation—including an assessment of OpenAI's own safeguards." Markey said that "building stronger defenses requires a full accounting of what goes wrong, and we cannot depend on companies with little incentive to disclose their failures to give us one" (Source: markey.senate.gov).

The bill's findings (section 2) state that AI "can strengthen cybersecurity controls while also increasing the volume and sophistication of cyber threats," that critical infrastructure sectors "such as clean and wastewater utilities and telecommunications networks" are susceptible to AI-enabled incidents, and that quantum computing and other emerging technologies are likely to "compound on artificial intelligence-driven risks" (S. 5541 — Cybersecurity and AI Board of Investigations Act of 2026 (bill text)).

It was introduced a day after Sens. Michael Bennet and Peter Welch announced the AI Regulator Act proposal, whose Federal Digital Commission would receive frontier-model critical safety incident reports within 15 days; the Markey bill instead creates an investigative body with no regulatory or enforcement role.

Scope and definitions

Section 3(a) takes the definition of artificial intelligence from section 238(g) of the National Defense Authorization Act for Fiscal Year 2019, and the terms cybersecurity threat, information system, security control and security vulnerability from section 2200 of the Homeland Security Act of 2002. An incident is an occurrence that actually or imminently jeopardizes, without lawful authority, the integrity, confidentiality or availability of information on an information system, or an information system itself. A near-miss is a circumstance in which a known or newly discovered security vulnerability is exploited in an attempt to jeopardize information on an information system, "through which no harm amounting to a significant cyber incident, as defined by the Board, is realized." Federal civilian executive branch agencies and systems exclude the Department of Defense, the intelligence community and national security systems (S. 5541 — Cybersecurity and AI Board of Investigations Act of 2026 (bill text)).

The board's review remit (section 3(b)(3)(B)) covers incidents affecting federal civilian or non-federal information systems, "particularly such systems that support critical infrastructure sectors"; near-misses "in which a significant cybersecurity incident was narrowly averted"; systemic vulnerabilities across sectors, vendors or technology categories, "including within the artificial intelligence sector"; and breakdowns in regulatory oversight, coordination or response. Its charter must limit eligible incidents to those for which all evidence, findings and recommendations are unclassified or declassified (S. 5541 — Cybersecurity and AI Board of Investigations Act of 2026 (bill text)).

Key provisions

Structure

The board is established as an independent establishment 90 days after enactment. Its five members are appointed by the President with Senate consent on the basis of technical qualification and knowledge of cybersecurity; no more than three may belong to one party; terms are five years, and the President may remove a member only for "inefficiency, neglect of duty, or malfeasance in office." Members are paid at no more than level IV of the Executive Schedule. The President designates a chair and vice chair, each for three years, with Senate consent; the chair is the board's chief executive and administrative officer, and must adopt a public charter with criteria for defining a significant cybersecurity incident and selecting incidents, and a conflict-of-interest system. The board meets at least four times a year, with a quorum of three (sections 3(c)–(d)). It is to establish an Office of Investigations, an Office of Trend Analysis and Learning, and an Office of Coordination with separate units for federal and private-sector liaison, and is to be supported by permanent federal staff that may include lead investigators, malware analysts, system engineers, legal counsel, digital forensic experts, and communications professionals and technical writers. It may also draw on federally funded research and development centers, special government employees and contracted private or academic investigators. It is exempt from the Federal Advisory Committee Act and the Government in the Sunshine Act (sections 3(b)(4), (f), (g)) (S. 5541 — Cybersecurity and AI Board of Investigations Act of 2026 (bill text)).

Reviews and reports

A review begins on a majority vote of all members, and the vote result is public. Each report must contain factual findings and actionable recommendations, a timeline, an analysis of contributing technical, organizational and systemic factors, a lessons-learned section identifying gaps in authorities or coordination, appendices with technical indicators of compromise and methodology, and a plain-language executive summary; it may contain only unclassified or declassified information. Reports go to the Senate Commerce and Homeland Security committees and the House Energy and Commerce and Homeland Security committees and are made public. A federal civilian agency to which a recommendation applies must respond in writing within 90 days, "even when the Federal civilian executive branch agency does not follow the recommended course of action" (section 3(e)). The board is to operate "independently from regulatory review and enforcement actions without assigning legal fault or liability" (section 3(b)(3)(H)) (S. 5541 — Cybersecurity and AI Board of Investigations Act of 2026 (bill text)).

Near-miss reporting and trend analysis

In coordination with the Director of the National Institute of Standards and Technology and the Assistant Secretary of Commerce for Communications and Information, the board must run a voluntary, anonymized near-miss reporting system for critical infrastructure sectors that protects reporters' identities and supports "proactive risk management and learning without attribution of fault or liability." It must also issue trend analyses on recurring threats "with attention paid to the novel and intensified risks relating to the continued development and adoption of artificial intelligence," and maintain a historical record of major incidents (sections 3(b)(3)(D), (F), (G) and 3(b)(5)) (S. 5541 — Cybersecurity and AI Board of Investigations Act of 2026 (bill text)).

Party participation

The chair may adopt a party system under which organizations affected by a matter under review, including providers of the implicated technology, supply factual input; such parties may not draft findings, shape recommendations or review a final report before publication (section 3(h)(2)) (S. 5541 — Cybersecurity and AI Board of Investigations Act of 2026 (bill text)).

Information gathering and protections

The chair may request documents and testimony from affected entities, responders, and anyone the board reasonably believes has relevant information. If a response is deficient, a subpoena may issue with the approval of a simple majority of members, after weighing whether the information can be obtained voluntarily and whether the subpoena would affect a regulatory, law enforcement, intelligence or diplomatic effort. Non-compliance is referred to the Attorney General for a civil enforcement action and may be punished as contempt. The subpoena power does not reach federal, state, local, tribal or territorial government entities or their employees acting in official capacities, and it does not override the Stored Communications Act. Subpoena authority may be delegated only to another board member, an electronically issued subpoena is invalid without the chair's or designee's cryptographic digital signature, and the board counts as a "governmental entity" for evidence-preservation requests under 18 U.S.C. 2703(f). Federal agencies must furnish information on request (section 3(i)) (S. 5541 — Cybersecurity and AI Board of Investigations Act of 2026 (bill text)).

Information provided voluntarily (including in response to a board request) is exempt from the Freedom of Information Act and state open-records laws, is treated as the provider's commercial, financial and proprietary information when so designated, waives no privilege (including trade-secret protection), is not subject to ex parte communication rules, may be used by the federal government only for listed cybersecurity, threat-response and law-enforcement purposes, and is protected from suit and from use as evidence or in discovery, except in investigating or prosecuting a person alleged to have committed the incident described. Subpoenaed material does not receive these protections but may be withheld from public disclosure at the board's discretion. All material must be protected at least to the FIPS 199 moderate-impact standard (sections 3(i)(3)(H), (j), (k)) (S. 5541 — Cybersecurity and AI Board of Investigations Act of 2026 (bill text)).

Drafting points

Two internal cross-references in the released text do not resolve: section 3(e)(1) refers to reviews "described in subsection (b)(2)(B)," which is the charter provision on trend analyses, where the review provision is subsection (b)(3)(B); and section 3(b)(3)(D) refers to the near-miss system "established under paragraph (4)," which establishes the board's offices, where the system is established in paragraph (5). The text contains no authorization of appropriations (S. 5541 — Cybersecurity and AI Board of Investigations Act of 2026 (bill text)).

Relationships

• instance-of: AI Incident Reporting
• related: Ed Markey
• related: AI Agent Containment Incidents, 2026 — the Hugging Face incident cited in the sponsor's release
• related: AI Regulator Act (Bennet–Welch proposal, 2026)
• related: Cybersecurity and Infrastructure Security Agency (CISA) — regulator role
• depends-on: S. 5541 — Cybersecurity and AI Board of Investigations Act of 2026 (bill text) — bill text
• related: National Institute of Standards and Technology (NIST) — coordinates the near-miss reporting system
• related: AI Regulator Act of 2026 — Section-by-Section Summary (Bennet and Welch)