Our newsroom AI policy

· media books ai · Source ↗

TLDR

  • Ars Technica published its formal AI editorial policy: humans write all stories, AI may assist research but cannot generate quotes, summaries attributed to sources, or documentary images.

Key Takeaways

  • AI tools approved for workflow may help reporters navigate large document volumes and search datasets, but output cannot be directly attributed to named sources.
  • Fabricated quotes, paraphrased AI summaries, and AI-extracted positions are all explicitly banned, regardless of how the material is labeled.
  • AI-generated images, audio, and video are prohibited as documentation of real events; synthetic media used in AI reporting must be labeled at point of use.
  • Every reporter using AI must disclose it to editors and retains full personal accountability – the tool is not a liability shield.
  • Policy was last updated April 22, 2026 and reflects standards already in force; publication is framed as reader transparency, not a new commitment.

Hacker News Comment Review

  • The policy is widely read as a reaction to a specific prior incident – Ars fired a reporter over fabricated quotes – making the no-attribution rule feel narrowly overfit rather than a general framework.
  • Commenters split on the research-assistance carve-out: permitting AI to summarize background documents while banning attribution of AI summaries is seen as a contradiction, since LLM summarization errors can still corrupt downstream reporting.
  • The visual media exception drew friction: allowing AI in thumbnail production while claiming “Ars is written by humans” reads as treating writing as uniquely human while offloading lower-status creative work to generation tools.

Notable Comments

  • @legitster: flags a systemic risk – AI trains on human-generated content, and if AI floods the web with synthetic output, the training corpus degrades, eroding the tools themselves.
  • @tantalor: notes the policy contains no mention of fact-checking, which is the more operationally relevant gap for AI-assisted research workflows.

Original | Discuss on HN