nxted
← Back to research
RLHFBy nxted Research Team· Published 29 May 2026· Updated 30 May 2026· 2 min read

RLHF in 2026: why quality concentration beats scale

High-profile breaches showed that black-box, low-context evaluation cannot scale safely. The alternative is concentrated, transparent expertise.

The wake-up call

The high-profile data breaches that hit AI evaluation platforms in 2026 were not outliers - they were predictable. When you build an AI evaluation platform on a US-housed, opaque, sub-contracted crowd, the failure modes compound.

The alternative

Quality concentration. Fewer evaluators, deeper expertise, full transparency on credentials and inter-rater agreement. EU/UK clients in particular cannot use platforms that hide who reviewed their AI.

What good looks like in 2026

  • Verified credentials, disclosed per project
  • Inter-rater agreement reported per project
  • Error taxonomies tied to the AI's deployment risks
  • A DPA on day one
n
nxted Research Team

Physical-AI data specialists at OFORO LTD (UK). We write about egocentric data, robotics dataset formats, RLHF and data governance. See what we build.