Full Text
Official text as published. Use Ctrl+F / Cmd+F to search within the document.
[Congressional Bills 119th Congress]
[From the U.S. Government Publishing Office]
[S. 5648 Introduced in Senate (IS)]
<DOC>
119th CONGRESS
2d Session
S. 5648
To direct agencies to be transparent when using automated and augmented
systems to interact with the public or make critical decisions, and for
other purposes.
_______________________________________________________________________
IN THE SENATE OF THE UNITED STATES
September 30, 2026
Mr. Peters (for himself and Mr. Lankford) introduced the following
bill; which was read twice and referred to the Committee on Homeland
Security and Governmental Affairs
_______________________________________________________________________
A BILL
To direct agencies to be transparent when using automated and augmented
systems to interact with the public or make critical decisions, and for
other purposes.
Be it enacted by the Senate and House of Representatives of the
United States of America in Congress assembled,
SECTION 1. SHORT TITLE.
This Act may be cited as the ``Transparent Automated Governance
Act'' or the ``TAG Act''.
SEC. 2. DEFINITIONS.
In this Act:
(1) Agency.--The term ``agency'' has the meaning given the
term in section 3502 of title 44, United States Code.
(2) Artificial intelligence.--The term ``artificial
intelligence''--
(A) has the meaning given the term in section 5002
of the National Artificial Intelligence Initiative Act
of 2020 (15 U.S.C. 9401); and
(B) includes the artificial systems and techniques
described in paragraphs (1) through (5) of section
238(g) of the John S. McCain National Defense
Authorization Act for Fiscal Year 2019 (10 U.S.C. note
prec. 4061; Public Law 115-232).
(3) Automated system.--The term ``automated system''--
(A) means a digital or computational system that
uses artificial intelligence techniques, including
machine learning, deep learning, neural networks, or
natural language processing, to autonomously generate
outputs, predictions, or recommendations that
substantially influence the outcome of critical
decisions, including computational processes that stand
alone or are embedded within another digital system or
application; and
(B) does not include--
(i) paper-based processes or physical
workflows; or
(ii) deterministic computational processes
or hard-coded logical algorithms that rate
without predictive, generative, or learning
capabilities, such as standard ``if-then''
decision trees.
(4) Critical decision.--The term ``critical decision''
means an agency determination, including the assignment of a
score or classification, related to the status, rights,
property, or wellbeing of specific individuals or groups, the
outcome of which meaningfully affects access to, or the cost,
terms, or availability of--
(A) education and vocational training;
(B) employment;
(C) essential utilities, including electricity,
heat, water, and internet;
(D) transportation;
(E) any benefits or assistance under any Federal
public assistance program or under any State or local
public assistance program financed in whole or in part
with Federal funds;
(F) financial services, including access to credit
or insurance;
(G) asylum and immigration services;
(H) healthcare;
(I) housing, lodging, or public accommodations;
(J) grants, contracts, and other legal agreements
with the Federal Government; and
(K) any other service, program, or opportunity
about which a determination would have a legal,
material, or significant effect on the life of an
individual, as determined by the Director.
(5) Director.--The term ``Director'' means the Director of
the Office of Management and Budget.
(6) Human-like service.--The term ``human-like service''
means dialogue or interaction with an individual through text,
voice, video, or other medium.
(7) Personally identifiable information.--The term
``personally identifiable information'' means information that
can reasonably be used to distinguish or trace the identity of
an individual (such as the name, social security number, or
biometric record of an individual), alone, or when combined
with other personal or identifying information that is linked
or linkable to a specific individual (such as date and place of
birth, or mother's maiden name).
(8) Plain language.--The term ``plain language'' means
language that is clear, concise, well-organized, and follows
other best practices appropriate to the subject or field and
intended audience, including an audience that may be disabled,
may not be proficient in English, or may otherwise be
disadvantaged or traditionally underserved.
(9) Prompt.--The term ``prompt'' means at a time not later
than the time of and at the place of--
(A) interaction with an automated system; and
(B) if applicable, issuance of an augmented
critical decision.
(10) Substantially influence.--The term ``substantially
influence'' means to serve as a principal basis for a critical
decision, including by producing an approval, rejection, score,
classification, or recommendation that an agency uses in whole
in part during an augmented critical decision process.
(11) Timely.--The term ``timely'' means within a period of
time in which the degree of reversibility is high enough to
avoid unintended outcomes created by artificial intelligence.
(12) Transparent automated government guidance.--The term
``transparent automated government guidance'' means the
guidance issued by the Director pursuant to section 3.
SEC. 3. TRANSPARENT AUTOMATED GOVERNANCE DIRECTIVES AND GUIDANCE.
(a) Directive Requirement.--Not later than 90 days after the date
of enactment of this Act, the Director shall issue a directive that--
(1) requires each agency publish, on or before the date
that is 180 days after the date of enactment of this Act, a
public inventory of all--
(A) augmented critical decision processes, and the
use of artificial intelligence within those processes,
including identification of any contracts and vendors
involved in those processes; and
(B) critical decisions to which this Act applies;
and
(2) requires each agency provide biannual notifications to
the public during the period in which the guidance required
under this section is being finalized.
(b) Part I Guidance on Critical Decisions.--The Director shall
issue transparent automated government guidance that includes--
(1) guidance for how an agency shall protect individuals'
privacy, civil rights, and civil liberties throughout augmented
critical decision processes;
(2) identification of any additional services, programs, or
other opportunities determined by the Director to be applicable
under section 2(4)(K);
(3) identification of critical decisions made by each
agency and guidance for how agencies should identify critical
decisions to which this Act would apply if the agency
implemented an augmented critical decision process to make
those critical decisions;
(4) examples of appropriate and inappropriate uses of
artificial intelligence in augmented critical decision
processes;
(5) with respect to augmented critical decision processes,
guidance for and examples of how agencies shall provide clear,
conspicuous, and prompt plain language notice to an individual
that a critical decision concerning the individual was made
using an augmented critical decision process, which shall
include information sufficient to enable individuals to appeal
and obtain recourse described in paragraph (7);
(6) the proper contents of the notice described in
paragraph (5) and examples of what the notice could look like
in practice, including a description of the role that
artificial intelligence played in that automated critical
decision process;
(7) a requirement that agencies establish a process for
individuals who are subject to adverse decisions in an
augmented critical decision process to appeal that adverse
decision and receive timely recourse, including publishing the
appeals process on an agency's website and establishing a
maximum number of days for the appeals process and a system
through which an individual can track their appeal request;
(8) with respect to automated systems that contribute to
augmented critical decision processes and interact with the
public, guidance for and examples of how agencies shall design,
develop, procure, or update those automated systems to comply
with this Act, including chatbot guidance in Part II under
subsection (c);
(9) with respect to critical decisions made by an augmented
critical decision process, guidance for how agencies, in
accordance with section 552a of title 5, United States Code,
should provide individuals--
(A) with the opportunity for an alternative review
by a human being working for or on behalf of the agency
with respect to the critical decision, independent of
the augmented critical decision process; and
(B) with an opportunity to request information
about the augmented critical decision process used in
making the critical decision, including identification
of the data, information, and process used to make the
decision;
(10) criteria for information and performance metrics that
each agency is required to track, collect, and make publicly
available relating to the augmented critical decision process
and any errors that occur during the augmented critical
decision process--
(A) to ensure that the information collected can be
used to determine whether each automated system and
augmented critical decision process covered by this Act
is accurate, reliable, fair and, to the greatest extent
practicable, explainable;
(B) to ensure that the agency follows a transparent
data retention, destruction, and storage security
protocol for the data that is collected for the
purposes of subparagraph (A); and
(C) to assist in enabling individuals to obtain
recourse described in paragraph (3);
(11) guidance for how agencies shall provide an opportunity
for public input, commensurate with the scale and impact of the
automated system--
(A) for any automated system used in an augmented
critical decision process that interacts with the
public; and
(B) for any new deployment of any automated system
used in an augmented critical decision process that
interacts with the public, prior to such deployment;
(12) guidance for how agencies shall establish ongoing
mechanisms to collect, analyze, and respond to feedback from
the public, especially individuals affected by automated
systems used in augmented critical decision processes,
including feedback arising from appeals, complaints, or user
experience; and
(13) guidance for how agencies shall periodically make
publicly available plain-language summaries describing material
issues raised through public feedback and the actions taken by
the agency in response, including to modify system design,
deployment, policy, or practice.
(c) Part II Guidance on Chatbots or Similar Technologies.--The
Director shall issue transparent automated government guidance that
includes--
(1) guidance on, including examples of, agency services and
public interactions which may be appropriate for using
artificial intelligence, such as chatbots, translation
services, fully automated voice agents, that mimic, simulate,
replicate, or replace services previously or regularly
conducted by humans, including interactions with agency
employees in person, on an agencies website, over the phone,
email, and mail and through text, voice, video, or other
medium;
(2) assessment, including examples, of which services and
public interactions may not be appropriate for using artificial
intelligence systems, such as chatbots, translation services,
fully automated voice agents, that mimic, simulate, replicate,
or replace services previously or regularly conducted by
humans, including interactions with agency employees in person,
on an agencies website, over the phone, email, and mail and
through text, voice, video, or other medium;
(3) guidance for how agencies shall provide prompt plain
language notices to individuals about the artificial
intelligence system that they are interacting with, including--
(A) a requirement that agencies provide clear,
conspicuous, and prompt plain language notice to
individuals, at the time and place that an individual
begins interacting with an artificial intelligence
system, such as a chatbot, translation service, or
fully automated voice agent that mimics, simulates,
replicates, or replaces services previously or
regularly conducted by humans, including on agency
websites, over the phone, email, and mail and through
text, voice, video, or other medium; and
(B) a requirement that during any extended
interaction with the automated system, a reiteration of
the actual notice under subparagraph (A) in a manner in
which a reasonable person can understand that they are
interacting with an automated system;
(4) guidance for how agencies shall establish a process by
which an individual interacting with an artificial intelligence
system described in this section can timely escalate requests
or address issues with a human being acting on behalf of the
agency to handle questions about relevant agency services;
(5) guidance for, and examples in practice of, how agencies
shall provide plain language notice to individuals about how
the agency will use any personally identifiable information
input into the artificial intelligence system, including a
notification that the agency will follow all data privacy laws
and that no personally identifiable information will be
transmitted to private companies, including a company that may
be operating the artificial intelligence system with which the
individual is interacting;
(6) performance metrics that each agency using an
artificial intelligence system for purposes described in this
section shall collect, monitor, and make publicly available,
including--
(A) metrics of the accuracy, efficiency,
effectiveness, cost, and accessibility of the
artificial intelligence system, including how relevant
metrics differ across language or other
characteristics;
(B) performance metrics by which the agency
measures successful delivery of that service and
customer experience, including metrics to quantify the
effectiveness, simplicity, speed, and transparency of
customer's interaction with the agency;
(C) a comparison of how the performance metrics
compare to metrics from prior month performance
measurements; and
(D) metrics indicating--
(i) the average number of agency
interactions per day;
(ii) the average time spent by an
individual interacting with the artificial
intelligence system;
(iii) the percentage of interactions
abandoned before the customer resolves their
concern, by language;
(iv) the percentage of issues resolved with
on interaction, by language; and
(v) the percentage of individuals who
abandon their interactions to seek help from a
customer service representative, by language;
(7) model contract terms an agency should use when
procuring a third-party automated system for purposes described
in paragraph (1), including terms and conditions that require--
(A) the agency to retain rights to Federal data and
any modifications to that data, including de-
identified, aggregated, or derived data generated in
performance of a contract;
(B) adequate pre-deployment testing, evaluation,
and ongoing monitoring of automated systems, including
performance, efficacy, safety, fairness, transparency,
accountability, appropriateness, lawfulness,
reliability, and trustworthiness;
(C) minimum standards for--
(i) security of data input into, processed
by, or output from the automated system;
(ii) data origin and data provenance,
retention, and disposal;
(iii) system design, development, testing,
operation; and
(iv) decommissioning of automated systems
and proper data management upon system or
contract termination;
(D) documentation to enable the agency to
understand the automated system's limitations, risk
mitigation measures, and cost to operate;
(E) disclosure of training data sources and any
known limitation of that training data that could
affect automated system performance; and
(F) incident reporting procedures to ensure timely
notification of system failures or security breaches;
and
(8) guidance for procurement officials involved in the
procurement process for automated systems, including
recommended trainings for such officials.
(d) Website for Templates, Guidance, Metrics.--The Director shall
publish all templates, links to all agency chatbots, performance
metrics, and the transparent automated government guidance on a new or
existing website of the Office of Management and Budget.
(e) Rule of Construction.--Nothing in this Act should be construed
to require less under or otherwise diminish any standard for the
protection of Federal data required under any other law or regulation.
(f) Prohibition on Commercial Use of Data.--The head of each agency
shall prohibit a contractor or third-party vendor of the agency from
using any data collected, generated, or accessed in the performance of
a contract, including de-identified, aggregated, or derived data
generated in performance of a contract, under this Act to--
(1) target, customize, personalize, or deliver
advertisements to any individual; or
(2) create, maintain, or augment a profile on any
individual for any purpose other than the direct performance of
the contract, including any changes to profiles that the
contractor or third-party vendor already holds or maintains on
an individual through commercial offerings.
(g) Account Creation.--The head of an agency may not require a
member of the public to create, sign up for, register, or establish an
account with a private or commercial entity, including any artificial
intelligence system or social media platform, as a condition of
interacting with the agency, accessing a service of the Federal
Government, or receiving any Federal benefits, including customer
support services.
(h) Account Separation.--If the head of an agency deploys a third-
party artificial intelligence system, including a chatbot, translation
service, fully automated voice agent, or any similar tool, in any
medium, including a website, telephone, email, mail, text, or video,
the head of the agency shall prohibit the agency or any operator,
developer, or deployer of the third-party artificial intelligence
system, including a contractor or vendor, from combining, cross-
referencing, or associating any data, metadata, or information derived
from the interactions of an individual with that artificial
intelligence system with any data, metadata, or information from a
commercial or personal account that the individual holds with the same
operator, developer, deployer, contractor, or vendor.
(i) Data Request.--
(1) In general.--If an agency deploys an interactive
artificial intelligence system (including a chatbot,
translation service, fully automated voice agent, or similar
tool) that communicates with members of the public, the agency
shall, upon request, provide to an individual interacting with
the artificial intelligence system a complete, unedited, and
machine-readable transcript or recording of the individual's
interactions with such interactive artificial intelligence
system.
(2) Timeline.--An agency providing a transcript or
recording to a requesting individual under paragraph (1)
shall--
(A) verify the identity of the requesting
individual to prevent unauthorized disclosure of
personally identifiable information; and
(B) provide the transcript or recording to the
verified individual not later than 3 business days
after receiving the request for the transcript or
recording.
(3) Self-service option.--An agency providing a transcript
or recording to a requesting individual under paragraph (1)
shall allow the requesting individual to also download or email
a transcript of their session with an interactive artificial
intelligence system directly from the interface immediately
upon conclusion of the interaction.
SEC. 4. AGENCY IMPLEMENTATION.
(a) Public Comment.--Not later than 180 days after the date of
enactment of this Act, the Director shall issue a preliminary version
of the transparent automated government guidance and make such version
available for public comment for a period of not less than 30 days.
(b) Agency Implementation of Transparent Automated Government
Guidance.--Not later than 180 days after the date on which the Director
issues the transparent automated government guidance under section 3 or
any subsequent update thereof, the head of each agency shall implement
the transparent automated government guidance or such subsequent
update.
(c) Comptroller General Report.--Not later than 2 years after the
date of enactment of this Act, and biannually thereafter, the
Comptroller General of the United States shall review agency compliance
with this Act and submit to the Committee on Homeland Security and
Governmental Affairs of the Senate and the Committee on Oversight and
Government Reform of the House of Representatives a report with
findings and recommendations.
SEC. 5. SUNSET.
Beginning on the date that is 10 years after the date on which the
Director issues the transparent automated government guidance, this Act
shall have no force or effect.
<all>