8 two-page summaries across the AI issues reaching constituents most. Each condenses a full, sourced CPAI research guide. Print this pack or save it as a single PDF to hand to an office.
A nonpartisan resource
The Center for Practical AI is a nonpartisan 501(c)(3) nonprofit. We provide education, research, and analysis, and we offer briefings and testimony on request. We do not endorse candidates or lobby for or against specific legislation. Everything here describes the evidence and the current landscape — the policy choices are yours.
Education & AI literacy
Safe AI Use
The same AI tool produces opposite outcomes depending on how it is used — and usage guidance is largely absent while adoption is near-universal.
What’s happening
AI assistants are now in daily use across work, school, and home. The evidence increasingly shows that the outcome — whether AI use builds capability or erodes it — depends less on how much people use it than on how.
What the evidence shows
In a randomized trial of about 1,000 students, unrestricted GPT-4 raised practice performance 48% but left those students 17% worse on a later unassisted exam; a version redesigned to give hints instead of answers eliminated the harm. A separate set of randomized studies found that people who used AI mainly for direct answers underperformed afterward, while those who used it for hints showed no deficit. One positive result, one negative — both randomized. The brief's credibility rests on presenting both.
Well-designed AI tutoring can roughly double learning gains over conventional instruction. The common thread across the evidence is design and usage mode, not access alone.
Where it reaches constituents
Students, workers, and families are adopting AI universally while usage guidance lags. Roughly three in four knowledge workers already use AI at work, and most received no training — so the habits that determine whether AI helps or harms are being formed by default, not design.
The current legal & regulatory landscape
The EU AI Act's Article 4 AI-literacy obligation has been in force since February 2025 — the first legal AI-literacy mandate. There is no comparable federal AI-literacy requirement in the United States. North Carolina's AI Strategic Roadmap commits to foundational AI-literacy training across all 100 counties by 2028.
Considerations policymakers are weighing
·How AI-literacy programs define "literacy" — tool mechanics versus the usage habits and evaluation skills that the evidence links to outcomes.
·Whether learning products disclose whether they are designed to give hints or answers.
·How programs measure outcomes (durable capability) rather than attendance.
AI use reshapes thinking in measurable ways — and the best-replicated finding is that people cannot feel it happening.
What’s happening
As AI takes over more cognitive work — drafting, analyzing, remembering — researchers are measuring effects on the underlying human capacities. The findings are real but modest, immediate-term, and easy to misjudge from the inside.
What the evidence shows
In a randomized trial, experienced developers were about 19% slower using AI on their own mature codebases — while believing they were 20% faster. That perception gap is the best-replicated part of this literature. Preregistered work with thousands of participants finds people systematically overestimate what AI saves them and reach for it even when it saves nothing.
Learning research adds the mechanism: the practice conditions that feel most fluent tend to produce the least durable learning. AI is very good at removing exactly the friction the learning was made of. Effects are small-to-moderate and short-term — stated honestly, this is a call for measurement, not alarm.
Where it reaches constituents
Every worker, student, and agency now relies on AI for cognitive tasks. Because the effects are invisible to self-perception, individuals and institutions cannot course-correct on feel alone — which makes measurement a public question, not just a personal one.
The current legal & regulatory landscape
This is largely unregulated territory. The relevant policy surface today is research funding and measurement standards rather than restriction; several major studies in this area are publicly and privately funded.
Considerations policymakers are weighing
·Support for longitudinal research on AI's cognitive effects, which barely exists.
·Measurement standards for "cognitive impact" claims, so products and programs can be compared honestly.
·Procurement questions schools and agencies can ask about how a tool is designed to be used.
Millions turn to AI in distress, and consumer chatbots carry no professional duty of care — the gap between a wellness product and a clinician.
What’s happening
People increasingly bring emotional and mental-health needs to general-purpose AI and to purpose-built companion apps. Some of these interactions reach people in crisis — and the tools they reach are not clinicians and are not regulated as such.
What the evidence shows
OpenAI has disclosed that roughly 0.15% of ChatGPT's very large weekly user base shows explicit indicators of suicidal planning or intent in a given week — over a million people (a company-reported, classifier-based estimate, attributed to OpenAI's own reporting). The American Psychological Association has advised that treating AI as a therapist or companion "carries real risk and is not advised."
The clinical evidence is narrow: therapy-chatbot studies show small-to-moderate short-term effects that fade by a few months, mostly for structured, purpose-built tools with clinician oversight — not consumer chatbots. Research on companion AI links intensive, self-disclosing, substitutive use to lower wellbeing, while also finding null effects at the average — a genuinely mixed picture.
Where it reaches constituents
Teens and adults in distress, and their families. Surveys find a majority of teens have used AI companions, and a meaningful share have turned to AI over people for serious conversations — often without knowing what the product is, or isn't, obligated to do.
The current legal & regulatory landscape
Character.AI litigation produced a May 2025 ruling treating a chatbot as a "product" for liability purposes, with settlements reported in January 2026 (terms undisclosed; no adjudicated causation). California's SB 243 (effective January 2026) sets disclosure, crisis-protocol, and minor-protection requirements for companion chatbots; Illinois restricts AI-delivered therapy; the FTC has opened a 6(b) study of companion chatbots (a study, not enforcement).
Considerations policymakers are weighing
·Disclosure requirements — whether a user is told they are not talking to a licensed professional.
·Crisis-protocol standards for what a product must do when a user signals danger.
·Age verification and minors' access.
·What clinical-validation evidence should be required before a product may claim to provide "therapy."
AI use at work has outrun training, and the fluency that helps the least-experienced most is reaching them last.
What’s happening
AI adoption at work has spread faster than almost any prior technology, while training has barely moved — leaving a gap between using AI and using it well. Meanwhile, the evidence on jobs is more nuanced than the headlines: mostly task change, with the first narrow employment signals only now emerging.
What the evidence shows
Company-reported surveys find about two-thirds of leaders would not hire someone without AI skills, while only 39% of AI users had received any training. The strongest randomized evidence shows AI raising productivity most for the least-experienced workers — about 35% in a field experiment with 5,000+ support agents. Employer-survey projections (WEF) estimate roughly 120 million workers are unlikely to receive the retraining they will need — a projection, not a measurement.
Where it reaches constituents
Workers at every level, and the employers and educators who train them. The equity finding — AI fluency helps the least-experienced most — means the training gap falls hardest on the workers with the least, unless programs reach them on purpose.
The current legal & regulatory landscape
The federal America's AI Action Plan (July 2025) includes worker-skilling programs, and Workforce Pell (effective July 2026) extends aid to short credential programs; the Digital Equity Act's capacity grants were terminated in May 2025. North Carolina's roadmap commits to credentialing 50,000+ residents in AI skills by 2028 — and notes the state cannot yet attribute layoffs to AI, because its reporting data was not built to capture it.
Considerations policymakers are weighing
·How workforce programs reach the automation-exposed and least-credentialed workers who gain the most.
·Whether "AI skills" funding measures durable capability or just enrollment.
·The data infrastructure needed to know what is actually happening to jobs.
A biased human decides one case. A biased system deployed statewide decides every case the same way — silently, and at scale.
What’s happening
Algorithmic decision-making now shapes outcomes in hiring, lending, healthcare, and the justice system. When these systems carry bias, the defining new feature is scale: one flawed model, applied to millions of decisions, in a way that is hard to see and can train the next system.
What the evidence shows
The documented cases are already sourced in CPAI's civil-rights guides: a risk-assessment tool that produced racially disparate false-positive rates; a health algorithm that under-referred Black patients because it used cost as a proxy for need; wrongful arrests from face recognition; a hiring tool scrapped after it penalized résumés containing the word "women's." The amplification point is concrete, not abstract: the same error, applied uniformly and invisibly across a whole population, with outputs that can become training data for the next model.
Where it reaches constituents
Anyone subject to an automated decision about a job, a loan, a benefit, or their liberty — disproportionately communities already least able to contest an opaque "the algorithm decided."
The current legal & regulatory landscape
New York City's Local Law 144 requires bias audits of automated hiring tools; Illinois and Colorado have enacted AI-in-employment and AI-accountability laws (Colorado's with a delayed effective date). In active litigation, Mobley v. Workday has let core AI-hiring discrimination claims proceed and conditionally certified a nationwide age-discrimination collective — with no finding of liability. The EEOC has issued and revised guidance in this area.
Considerations policymakers are weighing
·Impact-assessment and independent-audit requirements for consequential automated decisions.
·Transparency to affected individuals that an automated system was used.
·Appeal and human-review rights.
·Procurement standards for public-sector algorithms.
There is no comprehensive federal privacy law — so what happens to the data people put into AI depends on their zip code and the fine print.
What’s happening
People type sensitive information into AI tools that feel private but are company servers. That data can be used for training, retained, reviewed by humans, subpoenaed, breached, or sold — and the rules governing it are a patchwork.
What the evidence shows
In the New York Times copyright litigation, a federal court ordered OpenAI to hand over about 20 million de-identified consumer ChatGPT conversations in discovery — ordinary users' chats became evidence in a case they are not party to. "Anonymized" offers thin protection: a Nature Communications study estimated 99.98% of Americans could be re-identified from 15 demographic attributes. The FTC's amended COPPA Rule now requires separate parental consent before a child's data is disclosed to third parties, including for AI training.
Where it reaches constituents
Every person who uses an AI tool — and, acutely, children, patients, and anyone whose most sensitive data ends up in the least-protected place. The protection you get depends heavily on which state you live in.
The current legal & regulatory landscape
No comprehensive federal consumer-privacy statute exists; sector rules (HIPAA, FERPA, COPPA, GLBA) cover slices. About 20 states have comprehensive privacy laws in effect as of 2026; North Carolina does not. A newer layer of AI-specific state laws (Colorado, Texas, Illinois, California) is arriving alongside them.
Considerations policymakers are weighing
·Baseline data rights that do not depend on the state a person lives in.
·Rules for sensitive categories — genetic, biometric, health, and children's data.
·Transparency about training use, retention, and human review of what users enter.
People cannot reliably "spot the fake" — detection is near chance — so the durable protections are legal and procedural.
What’s happening
AI can generate convincing fake images and video of real people, and it is being weaponized — including as non-consensual intimate imagery targeting students, women, and children. The common advice to "look closely and spot the fake" does not match the evidence.
What the evidence shows
A meta-analysis of 56 studies found pooled human accuracy at detecting deepfakes was 55.5% — statistically close to chance. Stated neutrally: education promising people can spot the fake is not supported by evidence; process-based verification and legal remedies are the better bet.
Where it reaches constituents
Students and schools facing AI-generated NCII incidents; victims who need fast removal and support; and the broader public navigating synthetic media.
The current legal & regulatory landscape
The federal TAKE IT DOWN Act (signed May 2025) criminalizes non-consensual intimate images including AI "digital forgeries"; its platform notice-and-removal obligations took effect May 19, 2026, with a 48-hour removal window and FTC enforcement now underway. Most states also have non-consensual-imagery laws, with a newer wave addressing AI-generated deepfakes specifically.
Considerations policymakers are weighing
·Implementation oversight of platform notice-and-removal.
·School response protocols for incidents involving students.
·Support resources for victims.
·Investment in verification infrastructure rather than "spot the fake" education.
AI voice clones and synthetic identities are supercharging fraud against consumers — and older adults are losing the most.
What’s happening
Scammers now clone voices from seconds of audio, generate fake identification, and build convincing synthetic personas at scale. AI does not invent fraud — it lowers the cost and raises the believability, especially for impersonation and "emergency" scams targeting families and older adults.
What the evidence shows
The FBI's 2025 Internet Crime Report logged $20.9 billion in reported losses (up 26%) across more than a million complaints, and for the first time broke out a direct-AI category: 22,364 complaints totaling about $893 million. Adults 60+ reported roughly $7.7 billion, up 37%. The FTC reports a four-fold increase since 2020 in older adults losing $10,000 or more to impersonation scams. Because human deepfake detection is near chance, prevention rests on process — second-channel verification, a family safe word, and payment-method red flags.
Where it reaches constituents
Consumers and especially older adults and their families — the group reporting both the highest losses and the fastest growth. A judgment-based industry forecast projects AI-enabled fraud losses could reach $40 billion by 2027 (a projection, not a measurement).
The current legal & regulatory landscape
The FBI's IC3 provides the national reporting infrastructure; the FTC enforces impersonation rules. North Carolina's Executive Order 24 established a statewide AI Literacy and Fraud Prevention Training Program as state policy.
Considerations policymakers are weighing
·Reporting and data infrastructure so losses can be measured and traced.
·Bank and telecom verification protocols that add friction to high-risk transfers.
·Funding models for prevention education — and evaluation of what prevention actually works, since the canonical protective behaviors have not been rigorously tested.
The Center for Practical AI is a nonpartisan 501(c)(3) nonprofit. We provide education, research, and analysis, and we offer briefings and testimony on request. We do not endorse candidates or lobby for or against specific legislation. Everything here describes the evidence and the current landscape — the policy choices are yours.
