As of November 3, 2026
Issue areas that recur in hearings and briefings. Where groups disagree, positions are attributed.
- PrivacyPrompts can contain PII, case notes, or confidential records. Consumer tools may retain logs; enterprise contracts often set retention, residency, and non-training terms. Sector laws (health, education, children’s data, criminal justice) still apply when AI is only the interface. Cross-border transfers raise additional questions when a vendor’s subprocessors sit outside the buyer’s country. Privacy regulators in several jurisdictions have opened inquiries into scraping and model training; outcomes remain fact-specific.
- CopyrightNews publishers, authors, and other rightsholders have filed suits alleging unauthorized use of works in training corpora; developers argue fair use, implied license, or other defenses depending on the forum. As of Nov 3, 2026, major U.S. cases remain in litigation rather than settled into a single national statute. Separately, California’s AB 2013 requires certain generative-AI developers to document training-data categories (effective Jan 1, 2026).[30] Outputs that closely reproduce protected works raise distinct infringement questions from the training-data dispute. Public bodies that publish AI-assisted text still need clearance practices for third-party materials they paste into prompts.
- Deepfakes & electionsStates split between disclosure and prohibition models; some statutes face First Amendment challenges (NCSL, Oct 13, 2026).[28] Provenance standards such as C2PA Content Credentials record origin metadata; they do not by themselves detect unlabeled deepfakes.[47]
- EmploymentIllinois’s 2026 Human Rights Act amendment requires notice and bars AI use that has a discriminatory effect in covered decisions.[34] Labor organizations argue for bargaining over monitoring and automation; some employers argue AI assists productivity if human review remains.
- Critical infrastructureCISA and sector agencies publish guidance on AI-enabled cyber risk and on defending systems that themselves use AI. Failure modes include poisoned models, insecure agent permissions, and over-reliance on automated alerts.
- Government procurement & useOMB M-25-21 and M-25-22 set federal inventory, high-impact review, and acquisition practices.[19,20] Local examples include published AI inventories and staff guidelines (e.g., City of San José).[22,24]
- Child safetyTexas SB 20 and federal TAKE IT DOWN address AI-generated CSAM/NCII pathways.[15,16] Colorado’s Chatbot Safety Act (effective Jan 1, 2027) adds operator duties for conversational systems used by minors.[31] Schools and libraries face separate acceptable-use and student-data questions.
- LiabilityProduct, professional, and governmental liability doctrines are being tested in courts and legislatures. Agency policies commonly state that the human decision-maker remains responsible for official acts that use AI assistance.[19]