,,ii #aaith ,,congress #b;d ,session ,s4 #edga ,,in ,,the ,,senate ,,of ,,the ,,united ,,states ,september #bc1 #bjbf,mr4 ,coons 7for himself1 ,mrs4 ,britt1 ,mr4 ,schatz1 and ,mr4 ,lankford7 introduced the following bill2 which was read twice and referred to the ,committee on ,commerce1 ,science1 and ,transportation ,a ,,bill ,to require disclosure of certain information relating to artificial intelligence systems1 and for other purposes4 #a4 ,short title ,this ,act may be cited as the ,,ai ,systems ,transparency ,act4 #b4 ,definitions ,in this ,act3 7#a7 ,,ai red-teaming ,the term ,,ai red-teaming means a structured testing effort to find flaws and vulnerabilities of an ,,ai system1 often performed-- 7,a7 in a controlled environment and in collaboration with the developers of the ,,ai system2 and 7,b7 by dedicated teams that adopt adversarial methods to identify any flaws or vulnerabilities of the ,,ai system1 which may include harmful or discriminatory outputs1 unforeseen or undesirable system behaviors1 limitations1 and potential risks associated with misuse4 7#b7 ,artificial intelligence model ,the term artificial intelligence model means the part of an artificial intelligence system that can be adjusted to modify how the system behaves during inference4 7#c7 ,artificial intelligence system ,the term artificial intelligence system has the meaning given the term artificial intelligence in section #ejjb of the ,national ,artificial ,intelligence ,initiative ,act of #bjbj 7#ae ,u4,s4,c4 #idja74 7#d7 ,cloud provider ,the term cloud provider means an entity that offers on-demand network access to a shared pool of configurable computing resources1 such as networks1 servers1 storage1 and applications that can be rapidly provisioned and released with minimal management effort4 7#e7 ,commission ,the term ,commission means the ,federal ,trade ,commission4 7#f7 ,covered deployer ,the term covered deployer means an entity that operates a covered system which-- 7,a7 for the majority of the months in any #a-year period1 has1 or is expected to have1 not fewer than #a1jjj1jjj unique monthly users in the ,united ,states2 or 7,b7 allows minors to create accounts or profiles and has1 or is expected to have1 not fewer than #ajj1jjj unique monthly users in the ,united ,states4 7#g7 ,covered model ,the term covered model means a generative artificial intelligence model1 including an open weight model or an open source model1 that-- 7,a7 either-- 7i7 is accessible to the public through a website1 software application1 an application programming interface1 or a licensure through a cloud provider1 including as part of a broader ,,ai system2 or 7ii7 the model weights or architecture of which are made available to the public through the website of a developer1 download partners1 cloud partners1 or software development platform1 including in a situation in which accessing such model weights or architecture requires a user to create an account2 and 7,b7 is deployed in covered systems1 whether deployed by the developer or another deployer1 that-- 7i7 together have1 or are expected to have1 not fewer than #a1jjj1jjj unique monthly users in the ,united ,states for a majority of the months in any #a-year period2 or 7ii7 allow minors to create accounts or profiles and that together have1 or are expected to have1 not fewer than #ajj1jjj unique monthly users in the ,united ,states for a majority of the months in any #a-year period4 7#h7 ,covered platform ,the term covered platform means a website or software application used to interface with a covered system4 7#i7 ,covered system ,the term covered system means a covered model or collection of covered models1 together with any software component or infrastructure that-- 7,a7 are deployed as an integrated unit2 and 7,b7 operate in a coordinated manner to-- 7i7 handle different tasks or modalities2 or 7ii7 materially augment or enhance the capabilities of the covered model4 7#aj7 ,developer ,the term developer-- 7,a7 means an entity that initially builds1 designs1 codes1 produces1 pre-trains1 or owns an artificial intelligence system for internal use or for use by a third party2 and 7,b7 does not include an entity that is solely-- 7i7 a deployer of the artificial intelligence system2 or 7ii7 a substantial modifier of the artificial intelligence system4 7#aa7 ,generative artificial intelligence model ,the term generative artificial intelligence model means an artificial intelligence model that-- 7,a7 emulates the structure and characteristics of input data in order to generate derived synthetic data1 including images1 videos1 audio1 text1 code1 and other digital content2 7,b7 is trained on broad data2 7,c7 generally uses self-supervision2 7,d7 is designed for generality of output2 and 7,e7 is generally applicable across a wide range of contexts1 adaptable to a wide range of tasks1 or can issue a wide range of outputs in response to inferences4 7#ab7 ,machine learning ,the term machine learning has the meaning given the term in section #ejjb of the ,national ,artificial ,intelligence ,initiative ,act of #bjbj 7#ae ,u4,s4,c4 #idja74 7#ac7 ,minor ,the term minor means an individual that has not attained #ah years of age4 7#ad7 ,model inference ,the term model inference means the process of running data points through an ,,ai system to receive an output1 such as a number or text4 7#ae7 ,model weight ,the term model weight means a numerical parameter within an artificial intelligence model that helps determine the outputs of the model in response to inputs4 7#af7 ,open weight model ,the term open weight model means a machine learning model1 the weights of which are made publicly available1 allowing for access1 inspection1 modification1 and redistribution by any third party4 7#ag7 ,open source model ,the term open source model means a machine learning model1 the weight and model architecture1 training code1 and methodology of which are made publicly available1 allowing for a third party to fully reproduce1 audit1 and modify the model for custom applications4 7#ah7 ,post training and inference enhancements ,the term post training and inference enhancements means a process of adjusting or supplementing a pre-trained artificial intelligence model to improve performance on specific tasks1 better meet desired performance characteristics1 or adhere to safety constraints4 7#ai7 ,senior ,the term senior means an individual that has attained #fe years of age4 7#bj7 ,substantial modifier ,the term substantial modifier means an entity that significantly alters 7as defined by the ,commission7 the behavior or capabilities of a covered model through post training and inference enhancements4 7#ba7 ,synthetic data ,the term synthetic data means data that is artificially manufactured1 typically by a computer program4 7#bb7 ,vulnerable user ,the term vulnerable user means a minor1 a senior1 an individual with a disability1 or an individual experiencing a mental health crisis4 #c4 ,guidelines for model versioning and other designations 7a7 ,model versioning ,not later than #aej days after the date of enactment of this ,act1 the ,commission1 in consultation with the ,director of the ,national ,institute of ,standards and ,technology1 shall issue guidance regarding when a developer has produced a distinct generative artificial intelligence model through-- 7#a7 development1 including novel pre-training of a model architecture2 or 7#b7 any update to1 or improvement or iteration of1 an existing generative artificial intelligence model4 7b7 ,designation process 7#a7 ,in general ,not later than #aej days after the date of enactment of this ,act1 the ,commission1 in consultation with the ,director of the ,national ,institute of ,standards and ,technology1 shall promulgate regulations in accordance with section #eec of title #e1 ,united ,states ,code1 to establish criteria-- 7,a7 to determine when a deployment of a generative artificial intelligence model is expected to have sufficient users to meet the criteria described in-- 7i7 paragraph 7#f7 of section #b for a deployer to be considered a covered deployer2 or 7ii7 paragraph 7#g7 of section #b for a generative artificial intelligence model to be considered a covered model2 and 7,b7 to designate a generative artificial intelligence model or a particular deployment of a generative artificial intelligence model as a covered model or covered system1 based on a determination that it poses a significant risk to safety of minors1 vulnerable users1 or the public4 7#b7 ,requirements ,in establishing the criteria to make a designation under paragraph 7#a77;,b71 the ,commission1 in consultation with the ,director of the ,national ,institute of ,standards and ,technology1 shall do the following3 7,a7 ,consider the capabilities of the generative artificial intelligence model1 the contexts in which the model is deployed or reasonably likely to be deployed 7including the degree of access by vulnerable users71 and the severity and reversibility of the harms reasonably foreseeable from those capabilities4 7,b7 ,accompany any such designation with written findings1 which shall be made publicly available4 7,c7 ,provide the developer of the model or the deployer of the system with written notice and not fewer than #cj days to respond before any such designation takes effect1 except that a designation may take effect immediately upon a finding by the ,commission of imminent and significant risk1 in which case the ,commission shall provide such notice and opportunity to respond not later than #cj days after the effective date and shall confirm1 modify1 or rescind the designation not later than #fj days after the effective date4 #d4 ,artificial intelligence model disclosure mandates for developers 7a7 ,regulations 7#a7 ,in general ,not later than #aej days after the date of enactment of this ,act1 the ,commission1 in consultation with the ,director of the ,national ,institute of ,standards and ,technology1 shall promulgate regulations in accordance with section #eec of title #e1 ,united ,states ,code1 requiring each developer of a covered model 7including a distinct generative artificial intelligence model that meets the definition of a covered model7 to make publicly available the information described in subsections 7b7 and 7c7-- 7,a7 in a form determined by the ,commission that includes #b separate disclosures1 including-- 7i7 #a disclosure containing information relevant to1 and sufficiently detailed to facilitate clear understanding for1 a consumer2 and 7ii7 #a disclosure containing all information described in subsections 7b7 and 7c7 sufficiently detailed to permit meaningful review by an independent researcher2 and 7,b7 in a manner that-- 7i7 is accurate1 materially complete1 not misleading by reason of a material omission1 and adequate under standards determined by the ,commission2 7ii7 does not rely primarily on generalized or conclusory language2 7iii7 facilitates transparency1 public trust1 and independent evaluation2 7iv7 makes the information useful and actionable1 including that the information is-- 7,i7 reliably available in a clear and conspicuous section of the website of the developer that can be easily located and accessed by a consumer in alignment with the ,plain ,writing ,act of #bjaj 7#e ,u4,s4,c4 #cja note72 7,,ii7 searchable2 and 7,,iii7 available through an application programming interface2 7v7 does not disclose trade secrets2 and 7vi7 does not expose classified systems or create national security risks4 7#b7 ,consultation ,in promulgating the regulations under paragraph 7#a71 the ,commission and the ,director of the ,national ,institute of ,standards and ,technology shall engage relevant ,federal agencies 7including the ,department of ,health and ,human ,services1 the ,substance ,abuse and ,mental ,health ,services ,administration1 and the ,national ,institute of ,mental ,health7 and third-parties with expertise relating to consumer protection1 the safety of minors1 behavioral science1 mental health1 and human-computer interaction to ensure that the disclosure required by such regulations reflects best practices and includes the appropriate information4 7b7 ,initial public disclosure 7#a7 ,in general ,the information described in this subsection shall include the following3 7,a7 ,information relating to consumer transparency1 safety protocols for vulnerable users1 and other practices associated with the covered model1 including-- 7i7 an overview of preventative safety safeguards for minors implemented during pre-deployment testing and model development2 7ii7 an overview of measures taken to protect user privacy 7including the privacy of minors7 in training data1 such as through data filtering or anonymization methods1 and measures taken to protect user privacy 7including the privacy of minors7 on covered platforms2 7iii7 an overview of measures taken to ensure the safety of minors in training data1 such as through data filtering of child sexual abuse material2 7iv7 an overview of privacy standards pertaining to the management1 use1 storage1 and potential sale of user data 7including the data of minors71 including whether user data is used for targeted advertising1 user safety1 internal or external research1 and model training2 7v7 the durability and efficacy of vulnerable user safeguards 7including preventing the generation of child sexual abuse material and non-consensual intimate imagery71 as well as the resiliency and scope of such features2 7vi7 an overview of user and company reporting procedures for model behavior that poses a credible risk of material assistance to suicidal ideation or suicide planning1 or that poses a credible risk of material assistance to harm to others1 including model behavior learned in the post-deployment phase1 and whether such procedures were informed by crisis intervention experts and established standards2 and 7vii7 as appropriate1 the risk of negative impacts for vulnerable users on topics pertaining to mental health and well-being1 including disclosure of research on the short-term and long-term emotional and psychological impacts of use for vulnerable users1 and methodological standards for such research4 7,b7 ,system overviews of the covered model1 including-- 7i7 a high-level description of the intended1 unintended1 and prohibited uses of the covered model1 as reasonably anticipated by the developer2 7ii7 a high-level description1 including abilities and limitations1 of the covered model2 7iii7 if applicable1 a high-level overview of any difference between the covered model that is the subject of the disclosure and most recent previous version of the covered model that is considered a distinct generative artificial intelligence model according to the guidance issued under section #c2 7iv7 a list of each covered platform and interface through which the covered model is accessible2 7v7 a list of input and output formats supported by the covered model2 and 7vi7 an overview of the ability of the covered model to trigger actions or computations external to the covered model during model inference4 7,c7 ,an overview of the post training and inference enhancements of the covered model and relevant methodologies4 7,d7 ,information relating to safety1 security1 and evaluation protocols and practices associated with the covered model1 including-- 7i7 a high-level overview of procedures or guiding principles for-- 7,i7 determining what constitutes desirable model behavior2 7,,ii7 ensuring desirable covered model behavior2 and 7,,iii7 preventing undesirable covered model behavior2 7ii7 a description of methods or approaches used to evaluate and mitigate the risk that a model or the outputs of a related system are lower in quality or accuracy than intended use cases would require or expect2 7iii7 subject to subsection 7d71 an overview of measures taken to protect the covered model and training data from attack or misuse1 including-- 7,i7 traditional cybersecurity protocols1 such as access controls and multi-factor authentication2 and 7,,ii7 protections unique to ,,ai systems1 such as preventing model weight exfiltration1 data poisoning1 and prompt injections2 7iv7 with respect to open weight models1 an overview of the measures taken to protect training data from attack2 7v7 an overview of the criteria by which the developer chose to publicly release the covered model1 such as evaluation against a risk framework2 7vi7 a description of each of the evaluation criteria used to assess the risk profile of the covered model1 which shall address1 at a minimum1 chemical1 biological1 radiological1 and nuclear risk1 offensive cyber risk1 autonomous loss-of-control risk1 and any similar additional risk domain specified by the ,commission1 and which shall include1 for any such domain in which the developer conducted no evaluation1 an affirmative statement to that effect2 7vii7 an overview of the testing procedures and methods for achieving risk profile evaluations1 such as ,,ai red-teaming1 collaboration with external subject matter experts1 or solicitation of reports of model vulnerabilities from independent evaluators2 and 7viii7 an overview of the performance of the covered model on common public benchmarks determined by the ,commission in coordination with the ,director of the ,national ,institute of ,standards and ,technology1 including any related information and in a form necessary for independent researchers to verify the performance of the covered model on each benchmark1 such as prompts and configuration settings4 7#b7 ,timing of disclosure ,after the date on which the ,commission issues guidance under section #c1 a developer shall make publicly available the information described in paragraph 7#a7-- 7,a7 with respect to the most recent version of a covered model released to the public during the #b-year period before such date1 not later than #ij days after such date2 7,b7 with respect to a covered model released on or after such date1 concurrent with such release2 or 7,c7 with respect to the most recent version of a distinct generative artificial intelligence model released on or after such date that1 at a time after its release1 meets the definition of a covered model1 not later than #ij days after such time4 7c7 ,subsequent public disclosures 7#a7 ,requirement ,the information described in paragraph 7#b71 with respect to a covered model1 shall be made available to the public at such time1 as determined by the ,commission1 and on an ongoing basis until the ,commission determines that such disclosure is no longer necessary4 7#b7 ,information required ,the information described in this paragraph is as follows3 7,a7 ,any significant change 7as defined by the ,commission in coordination with the ,director of the ,national ,institute of ,standards and ,technology1 including any substantial modification to1 or observed differences from1 the most recent prior public disclosure under this section1 and not including any immaterial variation for which a disclosure requirement would overly burden platform operations7 to any of the information described in subsection 7b7 from the information reported in the initial information disclosure1 or1 if applicable1 any information reported in a subsequent disclosure4 7,b7 ,information including-- 7i7 data dictionaries or equivalent documentation that explain significant datasets1 as defined by the ,commission1 in the possession of a developer related to the covered model1 including datasets that the developer maintains relating to the users of the relevant covered platforms1 usage of the covered platforms1 content generated on covered platforms1 and actions taken to enforce the terms of service of covered platforms2 and 7ii7 information relating to usage of the covered model1 including-- 7,i7 an overview of the most common violations of covered model policies regarding acceptable use1 including common violations pertaining to minor safety2 7,,ii7 an overview of how often the model refuses a prohibited prompt and the developer has taken moderation actions in response to violations of acceptable use policies1 including what those actions are2 7,,iii7 an overview of information pertaining to attempts by users to utilize covered models to generate child sexual abuse material or non-consensual intimate imagery2 and 7,,iv7 an overview of known system capabilities for the utilization of the covered model to develop or advance cyber1 chemical1 biological1 radiological1 and nuclear threats4 7d7 ,special rules for disclosures 7#a7 ,in general ,with respect to any disclosure required under clause 7iii7 or 7vi7 of subsection 7b77#a77;,d7 or under subsection 7c77#b77;,b77ii71 such requirement shall not be interpreted to require a developer to disclose any information that1 as determined by the ,commission-- 7,a7 would compromise cybersecurity practices or protections or safety practices or protections of the developer2 or 7,b7 could provide material assistance to the development1 acquisition1 or deployment of chemical1 biological1 radiological1 nuclear1 or offensive cyber weapons or capabilities4 7#b7 ,rule of construction ,paragraph 7#a7 shall not be construed to permit a developer to withhold-- 7,a7 the fact that an evaluation in a risk domain described in subsection 7b77#a77;,d77vi7 was or was not conducted2 or 7,b7 a general characterization of the results of any such evaluation1 at a level of detail that does not include information described in paragraph 7#a74 7e7 ,reporting requirement for substantial modifiers ,not later than #a year after the date of enactment of this ,act1 the ,commission1 in consultation with the ,director of the ,national ,institute of ,standards and ,technology1 shall promulgate regulations in accordance with section #eec of title #e1 ,united ,states ,code1 requiring each substantial modifier to make publicly available the information described in subsections 7b7 and 7c71 as the ,commission determines appropriate with respect to a substantial modifier4 7f7 ,resources ,the ,commission shall provide covered platforms with technical assistance to support compliance with the requirements of this ,act1 particularly with respect to small and mid-sized platforms1 as determined by the ,commission4 #e4 ,mandatory disclosures for covered deployers of covered systems ,not later than #aej days after the date of enactment of this ,act1 the ,commission1 in consultation with the ,director of the ,national ,institute of ,standards and ,technology1 shall promulgate regulations in accordance with section #eec of title #e1 ,united ,states ,code1 to establish disclosure requirements for covered deployers with respect to covered systems consistent with the requirements for developers of covered models described in section #d1 as determined by the ,commission to be appropriate for a covered deployer with respect to a covered system4 #f4 ,enforcement 7a7 ,unfair or deceptive acts or practices ,a violation of section #d or #e or a regulation promulgated under this ,act shall be treated as a violation of a rule defining an unfair or deceptive act or practice under section #ah7;a77#a77;,b7 of the ,federal ,trade ,commission ,act 7#ae ,u4,s4,c4 #eg;a7a77#a77;,b774 7b7 ,powers of the ,commission 7#a7 ,in general ,the ,commission shall enforce this ,act in the same manner1 by the same means1 and with the same jurisdiction1 powers1 and duties as though all applicable terms and provisions of the ,federal ,trade ,commission ,act 7#ae ,u4,s4,c4 #da et seq47 were incorporated into and made a part of this ,act4 7#b7 ,privileges and immunities ,any person who violates section #d or #e or a regulation promulgated under this ,act shall be subject to the penalties and entitled to the privileges and immunities provided in the ,federal ,trade ,commission ,act 7#ae ,u4,s4,c4 #da et seq474 7#c7 ,authority preserved ,nothing in this ,act shall be construed to limit the authority of the ,commission under any other provision of law4 7#d7 ,rulemaking ,not later than #a year after the date of enactment of this ,act1 the ,commission shall promulgate in accordance with section #eec of title #e1 ,united ,states ,code1 such rules as may be necessary to carry out this ,act4 #g4 ,severability ,if any provision of this ,act1 or the application of such provision to any person or circumstance1 is held to be unconstitutional1 the remainder of this ,act1 and the application of the remaining provisions of this ,act to any person or circumstance1 shall not be affected4