A new bill in New York would require disclaimers on AI-generated news content

Original Article ↗ Hacker News Discussion ↗

New York is considering a law that would require news organizations to label content “substantially” created with generative AI and ensure human editorial review, aiming to protect press integrity and readers from undisclosed machine-written material. Commenters are sharply divided: some see labeling as a minimal safeguard comparable to food ingredient lists, while others predict Prop‑65‑style warning fatigue, selective enforcement, and easy evasion by bad actors that would leave only honest outlets burdened. Underneath the debate is a broader question of whether AI in journalism should be constrained through disclosure rules, stricter liability for inaccuracies, or not regulated at all given enforcement and free‑speech concerns.

Inevitability of AI vs. Role of Regulation

  • Some argue resistance to AI (disclaimers, bans) is emotional “status quo bias”; once a technology spreads, it can be regulated but not rolled back.
  • Others reject this fatalism, pointing out past social reforms (unions, rights, etc.) and insisting society can still shape AI’s use, especially in news.

Why Label AI-Generated News at All?

  • Concerns: AI news is often regurgitated, low‑value, and easy to weaponize for propaganda, fake reviews, political messaging, or deceptive ads.
  • News, in particular, should minimize “hallucinations” because misinformation cascades.
  • Some want all AI-generated content labeled, not just news; a few would prefer AI content banned entirely.
  • Others emphasize accountability: human editors and publishers should remain fully responsible for AI-assisted output.

Prop 65 Analogy and Overlabeling

  • Many predict a “California cancer warning” outcome: everything gets labeled “may contain AI,” users tune it out, and the signal becomes useless.
  • Overcompliance is expected because proving “no AI was used” is hard; risk‑averse organizations may label everything.
  • Counterarguments note Prop 65 did push companies away from toxic chemicals; labels can still shift behavior even if ubiquitous.

Enforcement, Detectability, and Abuse Risks

  • Technical detection of AI text is seen as inherently unreliable, especially as models improve and can mimic “human sloppiness.”
  • That implies laws will mostly bind honest actors; bad actors and foreign propagandists will ignore them.
  • Some fear selective or partisan enforcement (e.g., targeting disfavored outlets) and new litigation/trolling niches.
  • Others stress that many regulations (food safety, emissions, etc.) work via process audits and whistleblowers, not perfect detection.

Definitions, Edge Cases, and Scope

  • Major ambiguity: what counts as “substantially composed” by AI vs. AI-assisted (spellcheck, Photoshop, search, classifiers, summarizers)?
  • Worries that everything from camera filters to light AI editing will trigger labels, making them meaningless.
  • Some suggest tiered labels (AI-assisted vs AI-generated) or standards work (e.g., W3C disclosure schema).
  • There are First Amendment concerns about compelled speech; commercial vs. noncommercial content distinctions are debated.

Alternatives and Complements

  • Proposals include:
    • Labels for original reporting and explicit sourcing, independent of AI use.
    • Strong liability for misleading content regardless of whether AI was used.
    • User tools/filters to hide AI content and a possible market premium for “no-AI” journalism.