This bill requires the Attorney General and the Director of National Intelligence to jointly inventory and report on any artificial intelligence systems that have been given access to unminimized information collected under the Foreign Intelligence Surveillance Act. The report must describe each system’s function, developer, training data, testing and monitoring protocols, the scope of its access, and whether the Foreign Intelligence Surveillance Court (FISC) has addressed the system.
The itemized report must be delivered to congressional intelligence committees, the Judiciary Committees in both chambers, and the presiding judges of the FISC and FISC of Review; it must be provided in both classified and unclassified form, and the unclassified version must be posted publicly by DOJ and ODNI. The bill also requires that before any future AI system receives access to unminimized FISA information, DOJ and ODNI jointly notify those same oversight bodies and include an assessment of legality and fit with existing FISC procedures.
At a Glance
What It Does
The bill directs the Attorney General and Director of National Intelligence to submit a joint report within 120 days cataloging all AI systems that have access to unminimized FISA-derived information, including detailed technical and governance information for each system. It requires both classified and unclassified versions of the report and a public posting of the unclassified summary, and it creates a pre-notification and legal-compliance assessment requirement before any future AI is granted such access.
Who It Affects
Primary obligations fall on DOJ, ODNI, and intelligence components that operate or authorize AI tools with FISA access; private AI vendors and contractors whose systems are deployed will likely have to disclose training and provenance information. Congressional intelligence committees and FISC judges are formal recipients and oversight actors.
Why It Matters
This establishes the first statutory, cross-agency reporting requirement tying AI systems directly to FISA data access, creating a baseline for oversight and public transparency while also forcing agencies to document testing, human-review safeguards, and the provenance of models and datasets—topics with privacy, propriety, and operational-security implications.
More articles like this one.
A weekly email with all the latest developments on this topic.
What This Bill Actually Does
The bill defines “artificial intelligence” by reference to the National Artificial Intelligence Initiative Act of 2020 and then asks DOJ and ODNI to produce a single, joint accounting of every AI system that has had access to unminimized information collected under FISA. That accounting must describe what each system does, who built or trained it, what data trained or fine-tuned it, what testing and continuous-monitoring steps were taken, and what controls ensure ongoing human review and oversight.
The report must say when each system first accessed unminimized FISA data, what level of access it has, and whether the system aggregates other data sources during operation. The bill requires both a classified and an unclassified version; the unclassified version must be posted publicly on DOJ’s and ODNI’s websites.
Recipients of the report include the congressional intelligence committees, both Judiciary Committees, and the presiding judges of the FISC and the FISC of Review.Beyond retrospective disclosure, the bill creates a prospective gate: before any future AI system may touch unminimized FISA information, DOJ and ODNI must jointly notify the same oversight bodies, identify the proposed system and its intended access, and provide an assessment that the proposed use complies with FISA, applicable FISC procedures/opinions, and other laws and regulations. That assessment obligation imports legal review into the engineering and procurement path for surveillance tools.Practically, agencies will need internal inventories of deployed models, documented testing and monitoring procedures, and a capability to describe training datasets and model limitations at a level suitable for an unclassified public summary.
The requirement to identify developers, trainers, and entities that modified a model creates disclosure pressure on both in-house programs and private vendors; agencies will have to balance that with source-protection and proprietary-rights considerations.
The Five Things You Need to Know
The bill ties the statutory definition of “artificial intelligence” to section 5002 of the National Artificial Intelligence Initiative Act of 2020 (15 U.S.C. 9401).
DOJ and ODNI must submit the joint report not later than 120 days after the act’s enactment, in both classified and unclassified forms.
The report must identify the entity that developed, trained, initiated training of, or intentionally modified each AI system and must describe the model’s training or fine-tuning data and stated limitations.
The unclassified version of the report must be published publicly on the Department of Justice and Office of the Director of National Intelligence websites.
Before any future AI system receives access to unminimized FISA information, the bill requires DOJ and ODNI to jointly notify congressional intelligence committees, the Judiciary Committees, and the presiding FISC judges and to provide an assessment of legal compliance with FISA and applicable FISC procedures.
Section-by-Section Breakdown
Every bill we cover gets an analysis of its key sections.
Definition of artificial intelligence
The bill imports the existing statutory definition of “artificial intelligence” from section 5002 of the National Artificial Intelligence Initiative Act of 2020. Using an established definition avoids relitigating terminology but also fixes the scope of systems covered to the language and concepts in that 2020 statute, which matters for borderline or bespoke analytic tools.
Mandatory joint report on AI access to unminimized FISA information
Subsection (a) imposes a 120-day deadline for the Attorney General and the Director of National Intelligence to deliver a joint report listing every AI system that has accessed unminimized information collected under any FISA provision. Subsection (b) enumerates required contents: system name and purpose; testing, evaluation, and continuous-monitoring protocols including human-review mechanisms; developer/trainer identification; descriptions of training/fine-tuning data and model fitness; timing and scope of access and any additional data sources; whether prior notice was given and any FISC orders or opinions addressing the system; plus catch-all authority for AG/DNI to add other information they deem necessary. Subsections (c)–(e) identify recipients (congressional intelligence committees, both Judiciary Committees, and presiding FISC judges), require both classified and unclassified submissions, and mandate that the unclassified report be posted publicly by DOJ and ODNI.
Pre-notification and legal assessment before future AI access
Section 3 requires DOJ and ODNI to notify the same oversight bodies before any future AI system is permitted access to unminimized FISA-collected information. That notification must describe the proposed system and the nature of its access and include an assessment addressing whether the proposed use complies with FISA, FISC procedures or opinions, and other applicable law. This creates a formal legal sign-off and a paper trail prior to operational deployment, rather than relying solely on internal approvals.
This bill is one of many.
Codify tracks hundreds of bills on Privacy across all five countries.
Explore Privacy in Codify Search →Who Benefits and Who Bears the Cost
Every bill creates winners and losers. Here's who stands to gain and who bears the cost.
Who Benefits
- Congressional intelligence committees — gain a statutory, centralized inventory and standardized reporting to inform oversight, hearings, and legislative fixes.
- Presiding judges of the Foreign Intelligence Surveillance Court and Court of Review — receive formal notice and documented legal assessments about AI uses that could affect minimization and targeting procedures.
- Civil liberties organizations and the public — receive an unclassified account posted publicly, improving transparency into whether automated tools are processing raw FISA-derived content.
- Agency compliance and legal teams within DOJ/ODNI — receive a clearer procedural framework and a required process for legal assessments that can reduce later disputes about whether internal approvals were sufficient.
Who Bears the Cost
- Department of Justice and Office of the Director of National Intelligence — must compile technical inventories, assemble classified and redacted unclassified reports, and absorb staff time and analytic resources to meet the 120-day deadline and ongoing notification duties.
- Intelligence components and program offices that operate surveillance analytics — will need to document testing, monitoring, human-review protocols, and provenance of models and datasets, increasing program-management and legal-review burdens.
- Private AI developers and contractors whose systems are used — face pressure to disclose training datasets, model limitations, and other proprietary information when their tools are covered in the report or when agencies seek documentation for the required legal assessments.
- Classification authorities and program security offices — must adjudicate what stays classified, what can be redacted, and what can be included in the public summary, adding overhead and potential delays.
Key Issues
The Core Tension
The bill’s central dilemma is transparency versus secrecy: it demands public-facing transparency and robust oversight of AI tools that touch raw FISA information, but revealing too much—training data sources, model behavior, or operational parameters—risks exposing classified methods and vendor trade secrets and could undermine operational effectiveness; limiting disclosure protects sources and methods but undercuts the bill’s oversight and public-accountability goals.
The bill forces an inventory and a paper trail but leaves several implementation knots unresolved. First, the statutory cross-reference to the NAI Act definition avoids definitional fights but transfers ambiguity: many operational tools blend rule-based filters, statistical models, and ML components—determining which tools meet the statutory definition of “AI” will matter for coverage and may require additional agency guidance.
Second, the reporting prescription asks for training-data descriptions and developer identification while also demanding public posting of an unclassified summary. Agencies will need to reconcile disclosing model provenance with protecting classified sources, methods, and proprietary third-party trade secrets.
Expect redaction disputes and a potentially trimmed public report that may frustrate transparency advocates.
Operationally, the 120-day timeline is tight for a cross-agency inventory, particularly if agencies must trace historical access patterns, reconstruct testing records, and gather contractor-held documentation. The requirement that future uses be preceded by joint notice and a legal-compliance assessment creates a formal gate, but the bill does not spell out timelines for those notifications, standards for the adequacy of human-review mechanisms, or enforcement consequences if agencies fail to notify.
That ambiguity leaves room for differing interpretations about what level of detail suffices for oversight bodies and the FISC, and could produce uneven implementation across agencies.
Try it yourself.
Ask a question in plain English, or pick a topic below. Results in seconds.