← All Issues

AI: Promise, Risk & Accountability

Who controls artificial intelligence, who benefits from it, and who pays the cost — in jobs, energy, creative work, and existential risk?

Each issue breaks into the specific questions Congress actually fights over. Read each position, then head to the interactive version of this issue to mark which reflects your view and build a message to your representatives.

Component 1 of 5
Workforce displacement & retraining

Strong Regulation

AI is displacing workers at scale with no meaningful safety net, and Congress should require AI companies to help fund the transition. There's currently no dedicated safety net for workers displaced specifically by AI — existing programs like Trade Adjustment Assistance were built for trade dislocation, not automation. A federal workforce transition fund, paid into by AI companies and modeled on Trade Adjustment Assistance, could provide two years of fully funded retraining, wage insurance, and relocation support.

Balanced Oversight

The scale of AI-driven job displacement is still uncertain, but it's real enough to plan for now rather than after it peaks. A bipartisan AI Workforce Transition Fund, with corporate contributions scaled to revenue and tied to verified displacement metrics, could be phased in over five years rather than imposed all at once. Modeled on Trade Adjustment Assistance and the FUTURE Act (2019), it would fund community college retraining and apprenticeships without the overhead of a fully government-run program.

Industry Led

Government-managed retraining funds have a poor track record, and mandated levies on AI companies would just get passed on to consumers. Trade Adjustment Assistance, the closest existing model for a mandated retraining fund, has historically had low completion rates and high administrative overhead. Tax incentives for companies that voluntarily fund worker retraining are more efficient than a mandated levy, and avoid slowing the AI adoption that ultimately creates new jobs.

Documented compromise zone
Both parties have funded worker transition programs when triggered by documented, measurable displacement — TAA for trade, black lung benefits for coal miners. An AI Adjustment Fund structured as insurance, with contributions scaled to verified displacement data, avoids the "tech tax" framing while creating a durable mechanism. Start narrow: cover only industries where AI displacement is measurable; sunset unless renewed.
Trade Adjustment Assistance Act; FUTURE Act (2019, bipartisan); Workforce Innovation and Opportunity Act (WIOA)
Component 2 of 5
AI safety & existential risk

Strong Regulation

The risks from advanced AI are documented concerns raised by the researchers who build it, and Congress should require independent safety certification before deployment. Autonomous weapons and systems that could eventually outpace human control aren't fringe worries — they're raised by the researchers and labs building frontier AI themselves. A mandatory AI safety authority modeled on the FDA or FAA, with the power to certify high-risk systems before they're deployed, would put real teeth behind that concern.

Balanced Oversight

Advanced AI risk deserves serious policy, not panic or dismissal — and we already have a foundation to build on. The NIST AI Safety Institute and the NIST AI Risk Management Framework, along with the bipartisan Advancing American AI Act (2022), are a real starting point rather than a blank slate. Strengthening NIST's authority to certify AI in healthcare, critical infrastructure, and national security, paired with mandatory red-teaming and incident reporting, closes real gaps without building an entirely new regulatory structure.

Industry Led

Existential risk projections are speculative, and preemptive regulation risks ceding AI leadership to China without actually making anyone safer. Liability law — companies bearing real legal responsibility for the harms their systems cause — already creates strong market incentives for safety without government micromanagement of how models are built. Falling behind a less safety-conscious geopolitical rival in AI development is itself a serious risk, and heavy-handed regulation accelerates exactly that outcome.

Documented compromise zone
The sharpest bipartisan agreement is narrow but real: no AI in nuclear weapons targeting without human authorization; mandatory safety testing for AI deployed in healthcare and aviation. The NIST AI Risk Management Framework is already a bipartisan voluntary baseline — codifying it into law for high-risk sectors would give regulators authority without requiring Congress to define technical thresholds that will be obsolete in two years.
Advancing American AI Act (2022, bipartisan, signed into law); NIST AI Risk Management Framework (2023); Executive Order on AI Safety (Oct 2023)
Component 3 of 5
Energy & environmental impact

Strong Regulation

AI data centers are straining the grid and driving up household utility rates, and Congress should require them to run clean. Data centers are among the fastest-growing sources of electricity demand in the country, and that demand is showing up directly on household utility bills, not just in corporate energy budgets. Requiring 100% renewable energy for AI data centers by 2030, along with carbon disclosure tied to compute consumption, would put real limits on that growth's environmental cost.

Balanced Oversight

AI's energy footprint is real and growing fast, and disclosure plus efficiency standards can address it without dictating specific energy sources. The Department of Energy projects data center electricity demand could double by 2028, which is a real strain regulators need to plan for now, not after the fact. Requiring public disclosure of energy and water use per major AI model, and efficiency standards for new data center construction under DOE's existing authority, gets ahead of that without a mandate on where the power comes from.

Industry Led

Power costs are already the biggest data center expense, so companies have every incentive to improve efficiency without a mandate. AI is likely to be a net climate positive over time — it's already being used to optimize energy grids, accelerate battery and materials research, and improve climate modeling. Heavy-handed energy mandates risk pushing data centers offshore to countries with weaker environmental standards, which would make global emissions worse, not better.

Documented compromise zone
Disclosure requirements attract bipartisan support when the data serves a clear public interest. Mandatory energy and water use reporting for large AI model training runs — triggered by a compute threshold — is technically feasible, politically achievable, and does not mandate specific energy sources. It creates market pressure and informs grid planning without prescribing outcomes.
CHIPS and Science Act (2022) — domestic manufacturing efficiency model; DOE data center demand projections; American Energy Innovation Act
Component 4 of 5
Intellectual property & creative work

Strong Regulation

AI companies built billion-dollar systems on the life's work of artists, writers, and journalists without consent, credit, or pay, and Congress should require licensing. Active litigation like NYT v. OpenAI and Getty Images v. Stability AI shows creators are already fighting in court over training data used without permission. Requiring opt-in consent and royalty payments for copyrighted works used in commercial AI training, backed by a federal licensing registry, would settle that fight through legislation instead of years of case-by-case litigation.

Balanced Oversight

Copyright law needs modernizing for AI, not a wholesale rewrite — and the line between research use and commercial training is the right place to draw it. A workable framework treats transformative research use as fair use requiring no payment, while treating commercial training on copyrighted works as something that requires licensing. An optional federal registry where creators set their own licensing terms, paired with mandatory disclosure of training data categories and labeling of AI-generated content, protects creators without halting AI development.

Industry Led

AI learns from creative work the same way humans do, and copyright law has never required people to pay royalties for the books they study. Existing copyright law, applied through the courts in cases like NYT v. OpenAI, is already working through these questions without new legislation. New training-data restrictions risk driving AI development offshore to countries with fewer such requirements, harming U.S. competitiveness without actually protecting American creators.

Documented compromise zone
The Music Modernization Act (2018, bipartisan) solved an analogous dispute: streaming services were using songs without clear licensing, rights holders were undercompensated, and litigation was proliferating. Congress created a mechanical licensing collective with a blanket license and royalty distribution. The same model could work for AI — blanket training license with opt-out rights for creators, administered by an existing or new collective.
Music Modernization Act (2018, bipartisan); AI Labeling Act (introduced 2023); NYT v. OpenAI; Getty Images v. Stability AI
Component 5 of 5
Regulation & governance

Strong Regulation

AI already makes consequential decisions about housing, credit, healthcare, and parole with little transparency, and the push to preempt state AI laws would strip away the protections states have actually passed. States have enacted over 100 AI laws so far in 2026 covering things like algorithmic discrimination, chatbot safety for minors, and consumer protection — real protections built while Congress has yet to pass anything comparable. The federal preemption provisions in discussion-draft bills like the Great American Artificial Intelligence Act would wipe out most of that state-level progress for three years without putting an equivalent federal framework in its place.

Balanced Oversight

Congress is finally producing real, bipartisan AI governance drafts, and the fight now is over the shape of federal preemption, not whether federal rules should exist. The bipartisan Great American Artificial Intelligence Act discussion draft and the AI Labeling Act both show real cross-party interest in a federal framework covering transparency, model auditing, and content disclosure. But GAAIA's three-year preemption of state laws regulating AI model development is broad enough to sweep in laws well beyond the narrow federal-vs-state conflicts it's meant to resolve, which is where most of the actual disagreement now sits.

Industry Led

A patchwork of state AI laws is a real compliance burden, and the administration is right to push toward one clear federal standard. The White House's March 2026 National Policy Framework and the FTC's pending policy statement both argue that state laws forcing companies to alter the truthful outputs of AI models conflict with federal authority and should be preempted. With states enacting AI laws faster than in 2025, smaller AI companies increasingly face conflicting compliance obligations depending on which state their users happen to be in — a problem only federal preemption actually solves.

Documented compromise zone
The narrowest and most current point of bipartisan agreement isn't the broader regulatory-authority fight — it's disclosure. The AI Labeling Act of 2026, introduced by Senators Schatz (D-HI), Curtis (R-UT), and Warner (D-VA), would require large platforms to flag AI-generated audio, video, and image content, without taking a side on the harder preemption question dividing the GAAIA discussion draft. It suggests Congress can move on transparency requirements even while the state-preemption fight remains unresolved.
AI Labeling Act of 2026 (Sens. Schatz, Curtis, Warner, introduced June 25, 2026); Great American Artificial Intelligence Act discussion draft (Reps. Obernolte, Trahan, released June 4, 2026); White House National Policy Framework for Artificial Intelligence (March 20, 2026); Executive Order Ensuring a National Policy Framework for Artificial Intelligence (Dec. 11, 2025)
Take action on AI: Promise, Risk & Accountability