FOIA Cases
EPIC v. AI Commission
US District Court for the District of Columbia
Seeking public access to the records and meetings of the NSCAI
In EPIC v. AI Commission, EPIC filed suit to enforce the transparency obligations of the National Security Commission on Artificial Intelligence. The AI Commission, established by Congress in 2018 and chaired by former Google CEO Eric Schmidt, was charged with “review[ing] advances in artificial intelligence, related machine learning developments, and associated technologies” and making policy recommendations to Congress and the President. Yet for the first year of its existence, the Commission conducted its decisionmaking largely in secret. None of the Commission’s plenary or working group meetings were open to the public during that time, and few details of those meetings were contemporaneously disclosed.
Beginning in February 2019, EPIC submitted multiple open records and meetings requests to the Commission and Department of Defense under Freedom of Information Act (FOIA) and Federal Advisory Committee Act (FACA). After both agencies failed to comply with EPIC’s requests, EPIC brought suit on September 27, 2019. Although Judge Trevor N. McFadden denied EPIC’s motion for a preliminary injunction on October 16, the court ruled on December 3 that the AI Commission is an “agency” and must comply with the FOIA. On January 28, 2020, the AI Commission began producing records to EPIC pursuant to the FOIA. On February 28, the DOD also began producing records to EPIC.
On June 1, 2020, the court ruled that the AI Commission is subject to the Federal Advisory Committee Act in addition to the FOIA. The court rejected the Commission’s argument that it could not be subject to both statutes at once. “These different obligations are complementary, not conflicting,” Judge McFadden wrote. “[A]n entity subject to FOIA and FACA would need to look backward, producing records in response to requests, and forward, chronicling its activities and continually supplementing its records.” Accordingly, the court granted in part EPIC’s motion for summary judgment and ordered the Commission to “provide timely notice of its meetings, to open them to the public, and to make its records available for public inspection and copying” on a rolling basis.
On March 1, 2021—following a series of FACA-mandated public meetings—the AI Commission approved its final report. The report urges Congress and the President to implement key safeguards on federal AI deployment, including mandating AI impact and risk assessments, updating standards for Privacy Act notices and privacy impact assessments, establishing an independent auditor for AI systems, empowering the Privacy and Civil Liberties Oversight Board to conduct AI oversight, and establishing a task force to recommend legal restrictions on the use of AI. However, the report fails to propose any substantive limits on AI use for Congressional enactment, as EPIC urged in comments to the Commission. “Unless express, binding limits on the use of AI are established now, the technology will quickly outpace our collective ability to regulate it,” EPIC wrote. “The Commission cannot simply kick the can down the road, particularly when governments, civil society, and private sector actors have already laid extensive groundwork for the regulation of AI.” Controversially, the AI Commission’s final report also fails to endorse a ban on the use of autonomous weapons.
In October 2021, the Commission was disbanded following the expiration of its statutory mandate.
Background
The Privacy Risks Posed by the Use of AI
Artificial intelligence presents unique threats to privacy, human rights, and democratic institutions. The deployment of AI systems tests long-standing privacy safeguards governing the collection and use of personal data. For example, privacy laws require data minimization—the requirement that only necessary data be retained. Yet “[i]n the search for new connections and more precise analyses, it is tempting to give [a] system access to as much data as possible.” China, for instance, uses sophisticated AI surveillance technology to profile and control Muslim minority populations. Automated decisionmaking and profiling with AI systems can produce biased and inaccurate decisions, with serious consequences for the persons improperly targeted. Similarly, unrepresentative data sets can produce flawed AI models.
There is a clear need for human rights protections for AI systems in the national security context, where public oversight is often limited. Yet there are already indications that the U.S. Intelligence Community has failed to invest in vital AI safeguards. In May 2019, the Inspector General for the Intelligence Community highlighted a lack oversight for the use of AI, warning that “[i]nvestment asymmetry between mission performance and intelligence oversight in AI efforts could lead to an accountability deficit. . . . [T]here is little indication that investments in oversight of AI are currently a high priority.”
Privacy, security, and discrimination are not the only civil liberties and human rights issues raised by use of AI systems. International AI policy frameworks—including the OECD AI Principles (to which the United States is a signatory) and the Universal Guidelines for Artificial Intelligence—set out explicit rights and responsibilities concerning the use of AI systems. These include transparency and identification requirements, testing requirements, fairness, data quality, public safety, contestability, reliability, termination, and more.
Public Participation in AI Policymaking
The vast majority of AI policymaking around the world is conducted transparently and relies on public participation. National governments and international organizations routinely seek public input on AI policy. Europe’s High-Level Expert Group on Artificial Intelligence—a group of academic, civil society, and industry representatives—held a public consultation on Europe’s draft Ethics Guidelines for AI. The Council of Europe invited public comment on a draft recommendation concerning the human rights impacts of algorithmic systems. The Organization for Economic Cooperation and Development (“OECD”) established the Artificial Intelligence Group of Expert, representing OECD member organizations, held several meetings, sought comments from civil society, and produced the OECD Principles on Artificial Intelligence, which were endorsed by 42 nations and the G20.
Governments around the world have conducted transparent consultations on AI policy. Japan conducted a public consultation and published a draft AI research and development guidelines to prompt international debate over AI policymaking. The Australian government published proposed AI ethics framework for public consultation. And Canada and France made a joint public proposal for an international panel on artificial intelligence. All told, Australia, Canada, China, Denmark, EU Commission, Finland, France, Germany, India, Italy, Japan, Kenya, Malaysia, Mexico, New Zealand, Nordic-Baltic Region, Poland, Russia, Singapore, South Korea, Sweden, Taiwan, Tunisia, UAE, United Kingdom, and the U.S. have publicly released national AI strategies.
In the United States, EPIC—joined by leading scientific organizations and nearly 100 experts—filed a petition calling for public participation in federal efforts to develop AI policy. The coalition stated:
The reach of AI is so vast, so important, and encompasses so many issues, it is imperative that the Administration provide the American public the opportunity to comment on proposed policy initiatives impacting the American public. AI has the potential to improve our society, but only if proper policies are in place to provide the guidance needed to address the potential risks that accompany the potential benefits.
The National Science Foundation subsequently announced it would seek public comment on AI policy.
The National Institute of Standards and Technology published a plan for developing technical AI standards and sought public comments. The Office of Management and Budget solicited public comment about the use of federal data for AI research and development. And the President’s Executive Order on Maintaining American Leadership in Artificial Intelligence states that “[m]aintaining American leadership in AI requires a concerted effort to promote advancements in technology and innovation, while protecting American technology, economic and national security, civil liberties, privacy, and American values and enhancing international and industry collaboration with foreign partners and allies.”
The Formation and Structure of the AI Commission
Congress created the National Security Commission on Artificial Intelligence through the John S. McCain National Defense Authorization Act for Fiscal Year 2019 (“NDAA”), Pub. L. No. 115-232, § 1051, 132 Stat. 1636, 1962-65 (2018), signed into law on August 13, 2018. The NDAA directs the AI Commission “to review advances in artificial intelligence, related machine learning developments, and associated technologies.” The AI Commission is “an independent establishment of the Federal Government” that is “in the executive branch.” The AI Commission “shall be composed of 15 members” appointed “for the life of the Commission” by the Secretary of Defense, the Secretary of Commerce, and the chairs and ranking members of six congressional committees. The Commission “shall terminate on October 1, 2020.”
The Chairman of the Commission is Eric Schmidt, the former executive chairman of Alphabet Inc. and the former chairman and chief executive officer of Google Inc. The Vice Chairman of the Commission is Robert O. Work, former Deputy Secretary of Defense. As of December 2019, the Commission also includes:
- Safra Catz, chief executive officer of Oracle
- Steve Chien, supervisor of the Artificial Intelligence Group at Caltech’s Jet Propulsion Lab
- Mignon Clyburn, Open Society Foundation fellow and former FCC commissioner
- Chris Darby, chief executive officer of In-Q-Tel
- Ken Ford, chief executive officer of the Florida Institute for Human and Machine Cognition
- Jose-Marie Griffiths, president of Dakota State University
- Eric Horvitz, director of Microsoft Research Labs
- Andy Jassy, chief executive officer of Amazon Web Services
- Gilman Louie, partner at Alsop Louie Partners
- William Mark, director of SRI International’s Information and Computing Sciences Division
- Jason Matheny, director of the Center for Security and Emerging Technology and former Assistant Director of National Intelligence
- Katharina McFarland, consultant at Cypress International and former Assistant Secretary of Defense for Acquisition
- Andrew Moore, head of Google Cloud AI
Under the NDAA, the AI Commission is to “consider the methods and means necessary to advance the development of artificial intelligence, machine learning, and associated technologies by the United States to comprehensively address the national security and defense needs of the United States.” Congress has designated nine AI-related subjects for the Commission to review, including “ethical considerations related to artificial intelligence and machine learning as it will be used for future applications related to national security and defense.” Since it launched, the Commission has organized itself into four specialized working groups and has “decided to pursue Special Projects on three cross-cutting issues[.]” One of the Commission’s special projects concerns “the responsible and ethical use of AI for national security[.]”
EPIC’s FOIA Request and Lawsuit
In the first year of its existence, the AI Commission held dozens of plenary and working group meetings and received more than 100 briefings. The Commission met as a whole five times over that period: March 11, 2019; May 20, 2019; July 11, 2019; October 24, 2019; and January 15, 2020. None of these meetings were noticed in the Federal Register or open to the public, and the AI Commission published few details of those meetings contemporaneously.
Accordingly, EPIC filed two separate Freedom of Information Act/Federal Advisory Committee Act requests with the Department of Defense and the AI Commission. On Feb. 22, 2019, EPIC requested from the DOD:
On Sep. 11, 2019, EPIC requested from the AI Commission:
On September 27, 2019—following the failure of both the AI Commission and the DOD to expedite EPIC’s requests and failure to provide responsive records—EPIC filed suit against the AI Commission, Commission Chair Eric Schmidt, and the DOD in the U.S. District Court for the District of Columbia.
EPIC’s Interest
EPIC is one of the leading organizations in the country with respect to the privacy and human rights implications of AI use. In 2015, EPIC led an international campaign for “algorithmic transparency,” a practice which reduces bias and helps ensure fairness in automated decisionmaking. In 2018, EPIC led the drafting of the Universal Guidelines for Artificial Intelligence, a framework for AI governance based on the protection of human rights. The Universal Guidelines have been endorsed by more than 250 experts and 60 organizations in 40 countries. In 2019, EPIC published the EPIC AI Policy Sourcebook, the first compendium of AI policy frameworks and related AI resources.
EPIC regularly shares AI expertise and policy recommendations with Congressional committees, federal agencies, and international organizations. EPIC is currently seeking the release of a 2014 Department of Justice report to the White House concerning the use of predictive analytics and risk assessment algorithms in the criminal justice system. The DOJ has warned that assessments based on sociological and personal information rather than prior bad acts is “dangerous” and constitutionally suspect, citing the disparate impacts of risk assessments and the erosion of consistent sentencing.
Commission Documents
Initial Correspondence
- EPIC Letter Concerning Appointment of AI Commission Members (Dec. 19, 2018)
- EPIC Letter to AI Commission Members (Feb. 7, 2019)
- EPIC DOD FOIA & FACA Request (Feb. 22, 2019)
- EPIC AI Commission FOIA & FACA Request (Sep. 11, 2019)
- Acknowledgement Letter & Grant of Expedited Processing (Jan. 27, 2020)
First AI Commission Production (Jan. 27, 2020)
On January 27, 2020, the AI Commission produced 105 pages of records to EPIC. The records include an email exchange pertaining to a footnote in the Commission’s Interim Report (footnote 179) that sharply criticizes the European Union’s General Data Protection Regulation (GPDR). However, the AI Commission redacted the names of the parties to that email. The records produced also include NSCAI staff emails citing favorably to an article by former National Security Agency Counsel Glenn S. Gerstell.
Second AI Commission Production (Feb. 28, 2020)
On February 28, 2020, the AI Commission made a second production of records to EPIC. The records show that the AI Commission was previously aware of EPIC’s work on AI, including EPIC’s calls to eliminate algorithmic bias and ensure the transparency and accountability of AI systems. However, the Commission’s recent report to Congress did not endorse these requirements, instead criticizing the EU General Data Protection Regulation and calling for greater “government access to data on Americans.”
- Production Letter
- Index of Redactions
- Records Part 1
- Records Part 2
- Records Part 3
- Records Part 4
- Records Part 5
- Records Part 6
- Records Part 7
First DOD Production (Mar. 6, 2020)
On March 6, 2020, the Department of Defense made its first production of records to EPIC. The records disclosed concern the DOD’s role in the initial setup and funding of the AI Commission.
Third AI Commission Production (Mar. 31, 2020)
On March 31, 2020, the AI Commission made a third production of records to EPIC. The records produced include internal correspondence and a report about China’s social scoring, facial recognition tools, and AI-based surveillance. The internal report highlights the “draconian” consequences of China’s AI use but states that “Mass surveillance is a killer application” for AI and that “having streets carpeted with cameras is good infrastructure for smart cities[.]”
- Production Letter
- Index of Redactions
- Records Part 1
- Records Part 2
- Records Part 3
- Records Part 4
- Records Part 5
- Records Part 6
- Records Part 7
- Records Part 8
- Records Part 9
Fourth AI Commission Production (Apr. 30, 2020)
On April 30, 2020, the AI Commission made a fourth production of records to EPIC. The records produced provide the first public look at the AI Commission’s closed-door working group meetings.
- Production Letter
- Index of Redactions
- Records Part 1 (March 11, 2019 Plenary Meeting Materials)
- Records Part 2 (Organizational Documents)
- Records Part 3 (Memos and Status Reports)
- Records Part 4 (Working Group Materials)
- Records Part 5 (Outside Reports & Resources)
- Records Part 6 (May 20, 2019 Plenary Meeting Materials)
- Records Part 7 (Draft Preliminary Report)
- Records Part 8 (July 11, 2019 Plenary Meeting Materials)
Second DOD Production (May 8, 2020)
On May 8, 2020, the Department of Defense made a second production of records to EPIC. The records disclosed concern the DOD’s role in the initial setup and funding of the AI Commission.
Fifth AI Commission Production (May 29, 2020)
On May 29, 2020, the AI Commission made a fifth production of records to EPIC. The records produced include more details of the Commission’s closed-door working group meetings and a briefing about best practices for advisory commissions.
- Production Letter
- Index of Redactions
- Records Part 1 (Advisory Commission Best Practices Briefing)
- Records Part 2 (Working Group Materials)
- Records Part 3 (Memos and Status Reports)
- Records Part 4 (Outside Reports & Resources)
Sixth AI Commission Production (June 30, 2020)
On June 30, 2020, the AI Commission made a sixth production of records to EPIC. The records produced include more details of the Commission’s closed-door working group meetings and a report about public-private partnerships for the Commission’s July 11, 2019 plenary meeting.
- Production Letter
- Index of Redactions
- Records Part 1 (July 11, 2019 Plenary Meeting Materials)
- Records Part 2 (Working Group Materials)
- Records Part 3 (Outside Reports & Resources)
ODNI Production (July 23, 2020)
On July 23, 2020, the Office of the Director of National Intelligence made a production of records to EPIC that had been referred by the AI Commission. The production consisted of a presentation by Intelligence Advanced Research Projects Activity on “Artificial Intelligence and Threats.”
Seventh AI Commission Production (July 31, 2020)
On July 31, 2020, the AI Commission made a seventh production of records to EPIC. The records include a third-party presentation provided to the AI Commission about the use of “psychology and AI” to “help prepare an AI-enabled workforce.” The presentation endorses the use of AI job screening tools like HireVue and claims that “reducing time-to-hire is as important as making good decisions.” The presentation also argues that “sociometers can be used to train AI about effective communication” in the workplace. Sociometers are “wearable electronic device[s] capable of automatically measuring the amount of face-to-face interaction, conversational time, physical proximity to other people, and physical activity levels using social signals derived from vocal features, body motion, and relative location.”
- Production Letter
- Index of Redactions
- Records Part 1 (Presentation on AI in the Workplace)
- Records Part 2 (Working Group Agendas)
- Records Part 3 (Working Group Conclusions)
- Records Part 4 (AI Research Trends Report)
- Records Part 5 (AI Research Roadmap)
Eighth AI Commission Production (August 28, 2020)
On August 28, 2020, the AI Commission made an eighth production of records to EPIC. The records include a delegation letter from AI Commission chair and former Google CEO Eric Schmidt, as well as reports on AI research and “workforce automation.”
- Production Letter
- Index of Redactions
- Records Part 1 (Eric Schmidt Delegation Letter)
- Records Part 2 (AI Research Roadmap)
- Records Part 3 (GAO Materials)
- Records Part 4 (Commissioner Biographical Materials)
Ninth AI Commission Production (September 30, 2020)
On September 30, 2020, the AI Commission made a ninth production of records to EPIC.
- Production Letter
- Index of Redactions
- Records Part 1 (NSCAI & DOD Memos)
- Records Part 2 (Working Group Materials)
- Records Part 3 (Eric Schmidt Bio)
- Records Part 4 (Administrative Records)
Tenth AI Commission Production (October 30, 2020)
On October 30, 2020, the AI Commission made a tenth production of records to EPIC.
- Production Letter
- Index of Redactions
- Records Part 1 (Congressional Correspondence)
- Records Part 2 (Expert Bios)
- Records Part 3 (NSCAI Position Descriptions)
- Records Part 4 (Member Information)
- Records Part 5 (Administrative Records)
Eleventh AI Commission Production (November 30, 2020)
On November 30, 2020, the AI Commission made an eleventh production of records to EPIC. The records included emails from AI Commission members Safra Catz, Eric Horvitz, and Andrew Jassy. The records also revealed that Eric Schmidt, chair of the Commission and former CEO of Google, filed a 38-page ethics disclosure form.
- Production Letter
- Index of Redactions
- Records Part 1 (Safra Catz emails)
- Records Part 2 (Eric Horvitz emails)
- Records Part 3 (Andrew Jassy emails)
Twelfth AI Commission Production (December 31, 2020)
On December 31, 2020, the AI Commission made a twelfth production of records to EPIC. The records included emails from AI Commission members Andrew Moore and Eric Schmidt.
- Production Letter
- Index of Redactions
- Records Part 1 (Andrew Moore emails)
- Records Part 2 (Eric Schmidt emails)
Thirteenth AI Commission Production (January 29, 2021)
On January 29, 2021, the AI Commission made a thirteenth production of records to EPIC. The records consisted of emails from AI Commission chair Eric Schmidt.
- Production Letter
- Index of Redactions
- Records (Eric Schmidt emails)
Fourteenth AI Commission Production (February 26, 2021)
On February 26, 2021, the AI Commission made a fourteenth production of records to EPIC. The records consisted of emails from AI Commission members Eric Schmidt and Safra Catz, a report on enterprise risk management, and a memo from the Office of Management & Budget.
- Production Letter
- Index of Redactions
- Records Part 1 (Eric Schmidt emails)
- Records Part 2 (Safra Catz emails)
- Records Part 3 (ERM Report)
- Records Part 4 (OMB Memo)
OSTP Production (March 18, 2021)
On March 18, 2021, the Office of Science and Technology Policy made a production of records to EPIC that had been referred to the OSTP by the AI Commission. The records consisted of a redacted presentation about U.S. AI policy.
- Production Letter
- Records (AI policy presentation)
NSF Production (March 19, 2021)
On March 19, 2021, the National Science Foundation made a production of records to EPIC that had been referred to the NSF by the AI Commission. The records included information about the NSF’s work on AI.
- Production Letter
- Records Part 1 (AI policy presentation)
- Records Part 2 (AI policy presentation)
Legal Documents
EPIC v. NSCAI, No. 19-2906, __ F. Supp. 3d __ (D.D.C. Dec. 3, 2019)
- EPIC Complaint (Sep. 27, 2019)
- EPIC Motion for a Preliminary Injunction (Sep. 27, 2019)
- Government Opposition to EPIC’s Motion for a Preliminary Injunction (Oct. 9, 2019)
- EPIC Reply in Support of Motion for a Preliminary Injunction (Oct. 11, 2019)
- Transcript of Hearing on Motion for a Preliminary Injunction (Oct. 16, 2019)
- Order Denying Preliminary Injunction and Setting Expedited Briefing Schedule (Oct. 16, 2019)
- Government Motion to Dismiss FOIA Claims (Oct. 31, 2019)
- EPIC Opposition to Government Motion to Dismiss FOIA Claims (Nov. 7, 2019)
- Government Reply in Support of Motion to Dismiss FOIA Claims (Nov. 14, 2019)
- Opinion and Order Denying Government’s Motion to Dismiss (Dec. 3, 2019)
- EPIC Status Report (Dec. 19, 2019)
- Minute Order (Dec. 20, 2019)
- Government Answer to EPIC’s Complaint (Jan. 31, 2020)
- Government Motion to Dismiss FACA Claims (Jan. 31, 2020)
- EPIC Opposition & Motion for Partial Summary Judgment (Feb. 14, 2020)
- Government Opposition & Reply (Feb. 28, 2020)
- Joint Status Report (Mar. 3, 2020)
- EPIC Reply (Mar. 6, 2020)
- EPIC Notice of Clarification (Mar. 10, 2020)
- Memorandum Opinion (June 1, 2020)
EPIC Resources
- EPIC, EPIC AI Policy Sourcebook 2020 (2020)
- Comments of EPIC to the Council of Europe on Human Rights Impacts of Algorithmic Systems (Aug. 15, 2019)
- Comments of EPIC to OMB on Access to Federal Data for AI Research (Aug. 9, 2019)
- Comments of EPIC to NIST on AI Standards (May 31, 2019)
- Comments of EPIC to DOD on “Insider Threat Management and Analysis Center” (Apr. 22, 2019)
- Statement of EPIC to the Senate Comm. on the Judiciary on U.S. AI Policy (Nov. 30, 2018)
- The Public Voice, Universal Guidelines for Artificial Intelligence (Oct. 23, 2018)
- EPIC et al., Petition to OSTP for Request for Information on Artificial Intelligence Policy (July 4, 2018)
- EPIC v. DOJ, No. 18-5307 (seeking a DOJ report to the President and related records on the use of algorithms in the criminal justice system)
- EPIC, Algorithmic Transparency: End Secret Profiling (2015)
- EPIC et al., Petition for OSTP to Conduct Public Comment Process on Big Data and the Future of Privacy (Feb. 10, 2014)
News
- Court Finds Commission Subject to FOIA, Access Reports (Dec. 11, 2019)
- Joseph Ratliff, Judge Informs AI Commission It Is, In Fact, an “Agency”, MuckRock (Dec. 6, 2019)
- Eric White, Federal Court Decides AI Commission is Subject to FOIA, Federal News Network (Dec. 5, 2019)
- Daniel Wilson, EPIC Gets OK To Seek Federal AI Panel’s Records, Law360 (Dec. 4, 2019)
- Daniel Seiden, ‘Secret’ Artificial Intelligence Commission Subject to FOIA, Bloomberg Law (Dec. 4, 2019)
- Allan Blustein, Court Opinions Issued Dec. 3, 2019, FOIA Advisor (Dec. 4, 2019)
- Jeremy Kahn, A.I. Is Everywhere—But Where Is Human Judgment?, Fortune (Nov. 12, 2019)
- Andrew Kragie, Federal AI Panel Actions Must Be Public, EPIC Says, Law360 (Sep. 30, 2019)
- Alexandra S. Levine, New EPIC Lawsuit, POLITICO (Sep. 27, 2019)
- Kelsey D. Atherton, Why won’t the National Security Commission share its thoughts on AI?, C4ISRNET (July 15, 2019)
News
OMB Finalizes Guidance on Federal Government AI Procurement
October 4, 2024
OMB Finalizes Guidance on Federal Government AI Procurement
October 4, 2024
Support Our Work
EPIC's work is funded by the support of individuals like you, who allow us to continue to protect privacy, open government, and democratic values in the information age.
Donate