Artificial Intelligence Safety Board

(a) Artificial Intelligence Safety Board (1) Establishment (A) In general Not later than 30 days after the date of the enactment of this Act, the Secretary shall establish within the Department of Commerce a board to address artificial intelligence risks. (B) Designation The board established under subparagraph (A) shall be known as the Artificial Intelligence Safety Board (referred to in this Act as the Board). (C) Permanent status The Board shall be a permanent advisory committee, and section 1013 of title 5, United States Code, shall not apply to the Board. (2) Membership (A) Composition The Board shall be composed of members who are appointed as follows: (i) One member selected by the Director of the Institute. (ii) One member selected by the Secretary. (iii) One member selected by the Director of the Cybersecurity and Infrastructure Security Agency. (iv) One member selected by the Director of the National Security Agency. (v) One member selected by the Secretary of the Treasury. (B) Nongovernmental experts In addition to the members of the Board appointed under subparagraph (A), the Secretary shall appoint members who are not officers or employees of the Federal Government and who the Secretary selects from among individuals who— (i) are leading technical experts not affiliated with a developer or provider of artificial intelligence systems; (ii) are leading technical experts affiliated with developers or providers of artificial intelligence systems; (iii) are individuals with expertise in developing evaluations to test artificial intelligence systems; (iv) are individuals with expertise in consumer protection and antitrust jurisprudence; and (v) have knowledge or expertise that the Secretary determines would further the purpose of the duties of the Board. (C) International participation The Secretary, in coordination with and subject to the agreement of the Secretary of State, shall undertake negotiations with foreign partners, and enter into agreements as appropriate and subject to the disclosure requirements of section 112b of title 1, United States Code, to facilitate cooperative activities, regulatory reciprocity, and appropriate protection of intellectual property rights associated with addressing artificial intelligence safety and security risks, including— (i) the establishment of joint testing environments; (ii) jointly conducting competitive processes or competitive research programs to award cash prizes or other types of recognition for basic, advanced, and applied research, technology development, and prototype development that advance the security and safety of frontier artificial intelligence models; (iii) promulgating joint advisories or technical guidance concerning security or safety risks associated with frontier artificial intelligence systems; (iv) facilitating of secure information sharing regarding artificial intelligence security or safety incidents; and (v) standardizing evaluation protocols. (3) Terms and vacancies (A) Terms Each member of the Board shall serve for a term of not longer than 3 years and may be reappointed for 1 successive term. (B) Vacancy replacement The members of the Board shall develop a vacancy replacement procedure that includes— (i) for vacancies occurring due to the end of a member’s term, a vote not later than 90 days before the last day of the member’s term; and (ii) for vacancies occurring under subparagraph (C) or for any other reason, the chair of the Board shall nominate a replacement from the same stakeholder category under paragraph (2), to the extent practicable, as the member creating the vacancy, subject to approval by a majority vote of the members of the Board. (C) Removal A member shall be removed from the Board if— (i) the member fails to comply with the conflict of interest policy adopted pursuant to paragraph (5)(D); or (ii) the member is denied the requisite security clearance under paragraph (4)(A)(iii). (D) Chair The chair of the Board shall be selected by a majority vote among a quorum of the members appointed under paragraph (2) and shall serve not more than one 2-year term. (4) Member access to classified information (A) Access (i) In general Not later than 60 days after the date on which a member is first appointed to the Board and before the member is granted access to any classified information necessary to participate in a closed session pursuant to paragraph (5)(F), the Secretary shall determine, for the purposes of the Board, if the member should be restricted from reviewing, discussing, or possessing classified information. (ii) Management Access to classified information shall be managed in accordance with Executive Order 13526 (50 U.S.C. 3161 note; relating to classified national security information), or any subsequent corresponding Executive order. (iii) Clearance requirement (I) In general The Secretary shall sponsor each member of the Board for a security clearance at the Top Secret level with access to sensitive compartmented information, as appropriate, for the purposes of participating in carrying out the duties of the Board. (II) Denial Any member who fails to obtain a security clearance, including by being denied by the appropriate authorities or if the Secretary determines the member should be restricted from reviewing, discussing, or possessing classified information, shall be removed from the Board and a new member shall be appointed pursuant to the vacancy procedures under paragraph (3)(B). (B) Protection of information A member of the Board granted access to classified information shall sign and comply with the agreements to protect such information from unauthorized disclosure and to otherwise protect the classified information in accordance with the applicable requirements for the particular level of classification of the information. (C) Rule of construction Nothing in this paragraph shall be construed to affect the existing security clearance of a member of the Board or the authority of a Federal agency to provide or deny a member of the Board access to any specific pieces of classified information. (5) Procedures (A) Designated Federal officer The Secretary shall designate a Federal officer or employee to serve as the designated Federal officer of the Board, consistent with the requirements of chapter 10 of title 5, United States Code (commonly known as the Federal Advisory Committee Act). (B) Initial meeting and bylaws Not later than 60 days after the date of the enactment of this Act, the Board shall convene and establish bylaws that— (i) govern quorum and voting rules, including implementation of the decisionmaking majority voting requirement specified in paragraph (5)(C)(ii); and (ii) set deliverable timelines and meeting schedules. (C) Operating procedures Unless otherwise specified, the Board shall adopt written procedures governing its meetings, consistent with chapter 10 of title 5, United States Code, that include— (i) requirements for public notice of meetings and the maintenance of records and minutes; (ii) decisionmaking by majority vote of those present and voting; (iii) authorization for the establishment of subgroups as necessary, subject to the approval of the chair of the Board; and (iv) approval of the meeting agendas by the chair in consultation with the designated Federal officer under subparagraph (A) to ensure compliance with applicable laws. (D) Conflict-of-interest policy (i) In general The Board shall adopt and enforce a written conflict of interest policy to ensure that members of the Board have a fiduciary responsibility to the Board, a duty to report conflicts of interest, including the appearance of a conflict of interest, and do not participate in deliberations or votes from which they personally or their employer would directly and materially benefit. (ii) Required disclosures The policy under clause (i) shall require each member to publicly disclose all relevant financial and employment relationships, including relationships held by a spouse or immediate family member, and include recusal procedures in the event of a conflict. (iii) Records The designated Federal officer under subparagraph (A) shall maintain records of disclosures under clause (ii) of this subparagraph and make summaries of the disclosures available to the Secretary. (E) Threat information access The Director of National Intelligence, in coordination with the heads of other appropriate Federal entities, shall ensure that the Board has access to relevant intelligence, including through closed or classified briefings or the provision of classified information, when appropriate. (F) Closed sessions Notwithstanding section 1009 of title 5, United States Code, the Board may hold closed or restricted-access sessions when the Secretary determines that the matters to be discussed involve any of the following: (i) Classified information. (ii) An artificial intelligence security incident. (iii) An artificial intelligence security vulnerability. (iv) An artificial intelligence flaw. (v) Threat information. (vi) Proprietary business information. (vii) Other information exempt from public disclosure under section 552 of title 5, United States Code. (6) Duties (A) In general The Board shall— (i) develop a process to perform technical evaluations to determine what capabilities or combination of capabilities constitute high levels of performance at tasks that pose a serious risk to national security, national economic security, or public health or safety; (ii) develop processes and metrics for evaluating risks posed by frontier artificial intelligence models, model variants, checkpoints, or other artificial intelligence models or versions used during development, training, testing, evaluation, or red-teaming, including versions that are not publicly available, whether or not they are intended for eventual public release, where such models, variants, checkpoints, or other versions meet or could be modified to meet the risks described in section 2(3); (iii) develop technical standards and conformity assessment methodologies, including— (I) standardize formats and processes for publishing model or system cards with technical details of artificial intelligence systems; (II) recommendations for maintaining cybersecurity measures for developers or providers of artificial intelligence systems across the lifecycle; (III) processes and security controls for developers or providers of artificial intelligence systems to use to evaluate risks from employees or other personnel who have access to artificial intelligence systems developed or in development; and (IV) recommendations on appropriate financial and other resourcing for developers or providers of artificial intelligence systems, including small- and medium-sized developers or providers, to robustly engage in safety and security research focused on the deployment of frontier artificial intelligence models; and (iv) establish technical standards, security controls, and reference architectures for securing testing environments during evaluations of frontier artificial intelligence models, or models with known or reasonably foreseeable capabilities to discover and exploit software vulnerabilities without direct prompting by a human user, including— (I) procedures for effective risk-modeling prior to commencing any evaluation; (II) technical controls to ensure effective isolation and hardening of evaluation environments, including continuous re-evaluation of security configurations throughout evaluations; (III) policies and procedures for continuous monitoring of model behavior, including monitoring in real-time, using pre-established or conditional checkpoints, and through post-evaluation audits; (IV) policies, procedures, and technical controls for continuous monitoring of evaluation environments, including automated identification of changes to security controls or configurations; (V) policies, procedures, and technical controls for safeguarding identity and access management resources associated with the evaluation environment, or encompassing systems, from access by any model under evaluation; and (VI) conditions, policies, procedures, and technical controls for prompt termination of any evaluation in which a model has operated beyond the policies, procedures, or technical controls described in clauses (I) through (V) or that otherwise poses an imminent risk to any individual or property outside the scope of the evaluation. (B) Periodic reassessment of technical evaluations and best practices Not less frequently than once every year, the Board shall— (i) review each standard developed under subparagraph (A); (ii) determine whether the standard should be modified or revoked; and (iii) submit to the Secretary any proposed modification or revocation. (7) Support staff The Director of the Institute, acting through the Center for Artificial Intelligence Standards and Innovation (or any successor office or entity), shall provide staff and other support necessary to assist the Board in carrying out its duties under this Act. Such staff shall work under the direction of the Board, shall exercise on behalf of the Board the access provided to the Board under subsection (f), and shall remain employees of the Institute. (b) Adoption of standards (1) Deadline for submission of standards Not later than 90 days after the date of the establishment of the Board under subsection (a)(1), the Board shall develop and submit to the Secretary the proposed standards required under subsection (a)(6). (2) Adoption of standards (A) In general Not later than 30 days after the date on which the Secretary receives a proposed standard under paragraph (1), the Secretary shall adopt the proposed standard or the modified standards by rule. (B) Modification of standard The Secretary may modify a proposed standard received under paragraph (1) if the Secretary determines that the modification is necessary for the purpose of national security. (C) Publication The Secretary shall publish in the Federal Register the reasons for any modification made under this paragraph consistent with laws and regulations governing the disclosure of classified information or material. (c) Mandatory compliance Each developer shall comply with each standard adopted under subsection (b)(2) that is applicable to the developer. (d) Enforcement (1) In general The Secretary shall enforce compliance with this section. (2) Civil penalty A developer that violates subsection (c) shall be liable to the United States for a civil penalty of not more than $250,000 for each violation. (3) Continuing violations Each day during which a violation under subsection (c) continues shall constitute a separate violation. (4) Civil action The Attorney General may, at the request of the Secretary, bring a civil action in an appropriate district court of the United States— (A) to enjoin a violation of subsection (c); or (B) to recover a civil penalty imposed under paragraph (2). (e) Secure research-Test-Beds In developing a process to perform technical evaluations pursuant to section 6(A)(i), the Secretary may— (1) seek to utilize secure computing environments, on a reimbursable basis, provided by Federal partners, including— (A) the National Security Agency; and (B) the National Laboratories of the Department of Energy; and (2) make such secure computing environments available to private sector, Federal agencies, and qualified independent expert participants, on a cost-recovery basis, to engage in artificial intelligence security research, including through the secure provision of access in a secure environment for pre-deployment testing of any frontier artificial intelligence model prior to public release if security requirements necessitate hosting outside a commercial computing environment. (f) Requirement that developers of frontier artificial intelligence models give access to Board before public release Not later than 45 calendar days before a developer introduces into interstate or foreign commerce a frontier artificial intelligence model, the developer shall make available to the Board access to the frontier artificial intelligence model, including model’s weights, configuration files, runtimes, or software libraries necessary to operate the frontier artificial intelligence model.