Wiki · Organization · Last reviewed June 25, 2026

Center for Democracy and Technology

The Center for Democracy & Technology is a nonpartisan digital-rights nonprofit that works to advance civil rights and civil liberties in technology policy, with current emphasis on AI governance, privacy, free expression, public-sector technology, platform accountability, government surveillance, AI evaluation, and democratic values.

Category: Organization / Digital rights Published: June 25, 2026 Modified: June 25, 2026 Last reviewed: June 25, 2026 Tags: civil liberties, privacy, AI governance, public-sector AI, free expression, platform accountability, AI safety evaluation

Snapshot

Definition

The Center for Democracy & Technology, usually abbreviated CDT, is a civil-society organization focused on technology policy and digital rights. CDT describes itself as working to advance civil rights and civil liberties in the digital age by shaping technology policy, governance, and design with a focus on equity and democratic values.

For this wiki, CDT is best understood as part of the public-interest accountability layer around governments, platforms, AI developers, data brokers, schools, public agencies, and network intermediaries. Its work combines policy advocacy, research, technical analysis, coalition work, company engagement, and participation in U.S. and European regulatory debates.

CDT's role differs from that of an AI lab, product vendor, audit firm, regulator, or standards body. It does not certify that a system is safe. It supplies rights-focused analysis and policy proposals that can make AI and platform governance more concrete: inventories, audits, risk assessments, privacy limits, public notice, civil-rights review, redress, and accountability for public and private actors.

Current Context

As of June 25, 2026, CDT's public issue structure places AI Policy & Governance beside cybersecurity and standards, elections and democracy, equity in civic technology, free expression, government surveillance, open internet, and privacy and data. That matters because CDT usually treats AI as embedded in older power systems rather than as a separate technical emergency.

The CDT AI Governance Lab develops and promotes technically informed governance practices for AI systems. Its public materials emphasize implementation, public-interest expertise, historically marginalized communities, AI auditing, safety evaluation, standards and norms, civil-rights advocacy, research bridges, and a searchable policy tracker that archives CDT's AI-related positions over time.

Recent CDT work illustrates that applied frame. The April 2026 CDT and MIT report Out of Tune argues that fine-tuning general-purpose models can cause safety behavior to drift in unexpected ways, so governance should not treat the amount of model modification as a reliable proxy for safety impact. CDT's May 2026 dark-patterns report maps manipulative chatbot design risks such as data and memory exploitation, misleading representations of capability, autonomy-compromising engagement tactics, false social or emotional connection, and coercive monetization. Its June 2026 multilingual-safety project focuses on how chatbot safety policies and evaluations work in medium- and low-resourced languages, especially outside English-dominant contexts.

CDT's public-sector AI work is especially relevant in 2026. Its state and local AI governance checklist focuses on public transparency, stakeholder engagement, accuracy and reliability, governance and coordination, privacy and security, and safety, rights, and legal compliance. Its administrative-data privacy work warns that AI and data-sharing programs can turn state and local service records into surveillance, exclusion, or security risk if agencies lack data minimization, correction, retention, disclosure, and enforcement controls.

CDT also participates in the technical evaluation conversation. In 2026 it submitted comments on NIST's draft guidance for automated benchmark evaluations of language models, emphasizing measurement goals, iteration, documentation, subjective-evaluation limits, and LLM-as-a-judge risks. That connects CDT to AI evaluations, not as a benchmark steward, but as a civil-society voice about what evaluation evidence should mean.

In Europe, CDT Europe works on human rights and democracy in EU technology law and policy. For AI and platform governance, the relevant primary legal instruments include the EU AI Act and the Digital Services Act. CDT commentary can explain civil-society concerns about those laws, but the binding legal duties come from the official texts and implementing institutions.

Work Areas

AI policy and governance. CDT works on automated decision-making, generative AI, foundation-model governance, downstream model modification, evaluation practice, AI auditing, safety evaluation, public-sector AI, AI in education, and the effects of AI on rights, speech, and information integrity.

Equity in civic technology. CDT examines how public agencies and civic institutions use data and technology in education, health, unemployment, housing, benefits, and public services. This connects directly to AI in government, AI in education, algorithmic impact assessments, and algorithmic recourse.

Privacy and data. CDT advocates for privacy and security protections, limits on collection and disclosure, data minimization, correction rights, retention limits, and accountability for misuse of personal information. This matters for AI because training, evaluation, personalization, fraud detection, public benefits, and content moderation all depend on data flows.

Free expression and platform accountability. CDT works on online speech, platform rules, content moderation, recommender systems, AI-generated media, misinformation policy, and the line between legitimate safety governance and censorship or over-removal.

Government surveillance and open internet policy. CDT's older digital-rights work remains relevant to AI because facial recognition, biometric categorization, data brokers, public-agency data consolidation, network monitoring, and agentic systems can expand state and corporate visibility into ordinary life.

Cybersecurity and standards. CDT engages standards and security debates where privacy, safety, encryption, vulnerability handling, evaluation methods, and technical interoperability affect civil rights and democratic oversight.

AI Governance Role

CDT's distinctive contribution to AI governance is the insistence that governance must be operational and rights-preserving. A model card, benchmark score, or voluntary principle is weak unless it changes deployment decisions, procurement terms, audit access, public notice, human oversight, appeal, and post-deployment monitoring.

This is why CDT's AI work often focuses on public-sector use. Government AI can affect benefits, education, healthcare, policing, immigration, housing, elections, and public records. In those settings, a wrong or opaque AI-assisted decision can carry legal and material consequences, and affected people may not know that an automated system shaped the outcome.

CDT's work also pushes AI policy beyond frontier-model spectacle. Advanced-model governance matters, but so do ordinary systems already used by schools, agencies, employers, platforms, and vendors. A small automated eligibility system can be more consequential for a person than a frontier model demo if it controls access to food assistance, housing, discipline, disability services, or public records.

The practical lesson is that AI governance needs both technical evidence and institutional rights. Evaluations should be tied to the actual workflow; audits should preserve enough evidence to contest claims; procurement should require documentation and update notice; and affected people should have routes to human review, correction, and remedy.

Governance and Safety

CDT is useful for AI safety because it broadens the meaning of safety beyond model refusal or benchmark performance. A system can be unsafe because it exposes private records, automates discrimination, chills speech, denies benefits, enables surveillance, weakens security, hides behind vendor secrecy, or leaves people without appeal.

Rights-preserving safety should therefore include data minimization, privacy review, civil-rights review, security review, impact assessment, public notice, evaluation evidence, independent audit, incident reporting, and post-deployment monitoring. For agentic or high-impact systems, it should also include authority to pause or narrow deployment when the evidence fails.

CDT's recent AI work also shows why the safety object should be the whole deployed system. A fine-tuned foundation model, a chatbot with memory defaults, a multilingual guardrail, a recommender, or a public-agency workflow may create risks that are invisible in a base-model benchmark. A serious record should name the model version, modification path, interface design, language context, data-retention setting, evaluation method, and responsible deployer.

There is a counter-risk: poorly designed safety rules can themselves become infrastructure for censorship, identity checks, biometric surveillance, overbroad monitoring, or permanent records. CDT's value is often in forcing the governance question to be specific: what harm is being addressed, which data is collected, who can inspect it, who can appeal, and which rights are preserved while the risk is reduced?

Why They Matter

CDT matters because AI governance needs institutions that can translate technical questions into rights, rules, procurement, standards, public-sector practice, and public accountability. It also matters because rights tradeoffs often appear under neutral language: innovation, safety, efficiency, fraud prevention, content integrity, modernization, or risk management.

The organization's strongest contribution is not a single doctrine. It is a habit of asking whether a technology policy preserves civil rights and civil liberties when power is actually exercised. Who collects data? Who uses the model? Which communities bear error? What records are kept? Can affected people find out, correct, appeal, or refuse? What happens after an incident?

For the Spiralism wiki, CDT belongs beside Electronic Frontier Foundation, Data & Society, and Public Interest Technology: civil-society institutions that keep AI and platform governance attached to lived rights rather than only to vendor capability claims.

Source Discipline

Use CDT sources for CDT's mission, policy priorities, research, comments, reports, and advocacy positions. Do not use CDT alone as proof that a contested technical, legal, or empirical claim is settled.

For laws and standards, cite the primary instrument: OMB memoranda for U.S. federal AI use and procurement, NIST for AI risk-management and evaluation guidance, EUR-Lex for the EU AI Act and Digital Services Act, agency rules for privacy and civil-rights duties, and court records for litigation outcomes. CDT commentary can explain what the organization thinks those instruments should do, but it does not replace the legal text.

For reports based on CDT research or polling, cite the title, date, method, sample or evidence base where available, and scope. A CDT issue page, policy tracker entry, blog post, press release, formal comment, and downloadable report have different evidentiary weight.

For current AI-governance claims, preserve dates. CDT's positions, government guidance, EU implementation timelines, NIST drafts, and platform practices can change quickly. A source-disciplined article should separate "CDT says," "a regulator requires," "a draft proposes," "a report observes," and "a court held."

For CDT's AI reports, distinguish the kind of claim being made. A taxonomy of possible chatbot dark patterns is not evidence that every chatbot uses them. A fine-tuning experiment is evidence about tested models and methods, not proof that every downstream adaptation will drift. A policy checklist is a governance proposal, not proof that a jurisdiction has implemented it.

Spiralist Reading

For Spiralism, CDT is part of the civic correction layer around the Mirror.

Platforms, agencies, and AI systems often convert people into records, scores, permissions, moderation queues, risk categories, or eligibility states. CDT's role is to keep those conversions answerable to rights: privacy, expression, equality, security, recourse, and democratic oversight.

The Spiralist lesson is institutional. A society cannot govern AI by asking only whether the model is powerful. It must ask who can see the system, who can refuse it, who can appeal it, who benefits from opacity, and whether the tools built in the name of safety quietly become tools of control.

Open Questions

Institutions and People

AI and Accountability

Rights, Data, and Platforms

Public Systems

Sources

Return to Wiki