LIVE NEWSROOMToday's brief: The agentic stack consolidates around three runtimes
RSS
Thursday, May 14, 2026Vol. IV · No. 134
RabixAI
Intelligence on intelligence
Automated newsroom · updated hourly
SIGNALS
●GPT-5 Turbo pricing cut−42%●Anthropic ships Claude skills v2●Mistral Large 3 benchmark+11%●Cursor crosses 1M paid seats●Gemini context window10M tokens●EU AI Act enforcement begins●OpenAI launches o3 fine-tuning API●Meta open-sources Llama 4 base●xAI Grok 3 reasoning score+8%●GPT-5 Turbo pricing cut−42%●Anthropic ships Claude skills v2●Mistral Large 3 benchmark+11%●Cursor crosses 1M paid seats●Gemini context window10M tokens●EU AI Act enforcement begins●OpenAI launches o3 fine-tuning API●Meta open-sources Llama 4 base●xAI Grok 3 reasoning score+8%
Transparency

AI Use Disclosure

RabixAI is an AI newsroom — a publication that covers AI and uses AI to do its job. Both halves of that sentence carry obligations. This page documents exactly where AI is involved in our editorial process, what stays human, and what readers can expect to see labelled.

What AI does in our newsroom

We use AI tools — large language models, image generation models, embedding-based retrieval — at four stages of our editorial pipeline:

  • Signal ingestion. AI summarises and clusters news signals from RSS feeds, vendor announcements, and academic preprints, surfacing the events that warrant editorial review.
  • Drafting. AI generates an initial article draft from a sourcing packet (the source URLs, key facts, and editorial angle assigned by the assigning editor). Drafts are never published without human edit.
  • Fact-checking. AI cross-references claims in the draft against the cited sources and flags potential inconsistencies for the human fact-checker to resolve.
  • Illustration. Hero images for articles are generated using image models (currently FAL/Flux) under a fixed editorial style — flat geometric editorial illustration, restrained two-colour palette. Photographs of real people are never AI-generated.

What stays human

The following decisions are made by humans on every story:

  • What gets covered and what doesn't.
  • The editorial framing — angle, headline, lede, conclusion.
  • Sourcing decisions — which sources are reliable, which need corroboration, which are excluded.
  • Fact-check resolutions — when an AI flag is raised, a human verifies and decides.
  • Final publish/hold decision — every article clears the human editor before it becomes indexable.
  • Corrections — every correction is decided and written by a human editor.

Our publishing pipeline enforces this. The pipeline ends at status needs_review; only an explicit human-triggered action via our admin desk transitions an article to indexable.

What AI never does

  • Quotes. AI does not invent quotes. Quotes in RabixAI articles come from named sources, transcripts of public statements, or written correspondence — never paraphrased and attributed without verification.
  • Statistics. Every numerical claim is verified against a primary source. AI-generated statistics that cannot be traced are removed.
  • Photographs of real people. We never use AI image models to generate photographic-style images of identifiable individuals. Editorial illustrations are flat, geometric, and obviously stylised.
  • Final publication. No article publishes automatically. A human editor holds the final approval.

Models used

For full transparency, the AI models currently in use across our editorial pipeline:

  • Drafting and editorial agents — Anthropic Claude Sonnet 4.6 and Claude Opus 4.7.
  • Research and signal classification — Groq-hosted Llama 3.3 and DeepSeek for cost-sensitive workloads.
  • Image generation — FAL.ai Flux/Schnell for hero images, with a fixed RabixAI house style.

Model selection changes as we evaluate new releases. We disclose any model change that materially affects output quality.

Labelling

Every published article carries a "VERIFIED · EDITED BY HUMANS · SOURCES VERIFIED" trust ribbon below the deck. This is not decoration — it's a statement that the article cleared our publish checklist (see the methodology) including human edit and fact-check.

Hero images are stylised illustrations, not photographs. We do not run a separate "AI-generated" label on illustration because every illustration in our editorial system is AI-assisted under a fixed house style — labelling each one would be redundant.

If something feels off

If an article reads like it wasn't edited by a human, or makes a claim you can't trace to a source, please tell us. Email corrections@rabixai.com with the URL and the specific concern. AI-assisted journalism is new enough that catching its failure modes is part of how we improve.