- Record: Senate Floor
- Section type: Amendments
- Chamber: Senate
- Date: June 24, 2026
- Congress: 119th Congress
- Why this source matters: This section came from the Senate floor portion of the record.
SA 6154. Mr. WYDEN submitted an amendment intended to be proposed by him to the bill S. 4784, to authorize appropriations for fiscal year 2027 for military activities of the Department of Defense, for military construction, and for defense activities of the Department of Energy, to prescribe military personnel strengths for such fiscal year, and for other purposes; which was ordered to lie on the table; as follows:
At the appropriate place in subtitle B of title XV, insert
the following:
SEC. 15. ASSESSMENT OF DATA QUALITY AND GOVERNANCE
PRACTICES FOR TARGET IDENTIFICATION ARTIFICIAL
INTELLIGENCE SYSTEMS.
(a) In General.—Not later than December 31, 2027, the
Secretary of Defense shall complete a comprehensive
assessment of—
(1) the quality of data and the potential for harmful bias
and hallucinations in artificial intelligence systems used
for target identification, sensor processing, and decision-
making support;
(2) the risks to mission effectiveness from automation bias
and hallucinations by such artificial intelligence systems;
and
(3) the risks of adversarial manipulation of training data
and other data inputs used by such artificial intelligence
systems.
(b) Contents.—
(1) Data.—The assessment required by subsection (a)(1)
shall include an assessment of data used to train—
(A) target identification tools, including tools used for
the Maven Smart System;
(B) intelligence, surveillance, and reconnaissance systems;
and
(C) weapon systems that have lethal, offensive strike
capabilities that are autonomous or planned to become
autonomous; and
(D) weapon systems subject to senior review under
Department of Defense Directive 3000.09.
(2) Automation bias and hallucinations.—The assessment
required by subsection (a)(2) shall include an assessment
of—
(A) whether and how the intermediation of artificial
intelligence systems in human decision-making leads to over-
reliance or harmful deference to artificial intelligence
system recommendations; and
(B) the impact of erroneous or misleading outputs,
including hallucinations on mission effectiveness and the
appropriateness of the integration of artificial intelligence
systems into target identification and decision-making
support.
(3) Adversarial manipulation.—The assessment required by
subsection (a)(3) shall include an assessment of whether
artificial intelligence systems have been tested against
adversarial manipulation, including through data poisoning
and backdoor attacks through embedded hidden malicious data,
under operationally realistic conditions.
(c) Briefing.—Not later than February 1, 2027, the
Secretary shall brief the appropriate congressional
committees on the completed assessment required by subsection
(a) and recommendations how to improve the quality of the
assessed data.
(d) Definitions.—In this section:
(1) Appropriate congressional committees.—The term
“appropriate congressional committees” means—
(A) the Committee on Armed Services and the Select
Committee on Intelligence of the Senate; and
(B) the Committee on Armed Services and the Permanent
Select Committee on Intelligence of the House of
Representatives.
(2) Artificial intelligence system.—The term “artificial
intelligence system” has the meaning given the term
“artificial intelligence” in section 5002 of the National
Artificial Intelligence Initiative Act of 2020 (15 U.S.C.
9401).
(3) Autonomous; planned to become autonomous.—
(A) Autonomous.—The term “autonomous”, with respect to a
weapon system, means that the weapon system, once activated,
can select and engage targets without further intervention by
an operator, as defined in Department of Defense Directive
3000.09; or
(B) Planned to become autonomous.—The term “planned to
become autonomous”, with respect to a weapon system, means
that the weapon system has the potential to be deployed in a
manner that would qualify as an autonomous weapon system
under Department of Defense Directive 3000.09.
(4) Hallucination.—The term “hallucination” means an
output produced by an artificial intelligence system that
contains factually incorrect or wholly fabricated
information, including information that a person may find
convincing or credible.
(5) Quality of data.—The term “quality of data”
includes—
(A) the accuracy of data labeling;
(B) the condition of the data, including an completeness;
(C) the accuracy of data indexing;
(D) the suitability of the data for the intended task;
(E) the freedom of the data from unintended bias; and
(F) variance in quality of all training across all data
used to train an artificial intelligence system.