Artificial Intelligence

OpenAI Launches Embedded Evaluator Program Amid Researcher Skepticism

· GlobeWire Daily Editorial Desk · Source: www.npr.org

2 min read

Key takeawaysOpenAI has officially initiated a new program inviting researchers from four external organizations to serve as "embedded assessors" within the company. The initiative, announced by CEO Sam Altman, aims to provide greater transparency regarding how the company evaluates and manages risks associated with its AI models. While OpenAI frames this as a…
OpenAI Launches Embedded Evaluator Program Amid Researcher Skepticism

Original GlobeWire Daily editorial illustration — not an event photograph.

OpenAI has officially initiated a new program inviting researchers from four external organizations to serve as "embedded assessors" within the company. The initiative, announced by CEO Sam Altman, aims to provide greater transparency regarding how the company evaluates and manages risks associated with its AI models. While OpenAI frames this as a significant step toward third-party safety oversight, several independent researchers have expressed concern that the program lacks the depth and access originally promised.

Program Scope and Access

According to OpenAI, the embedded assessors will have access to specific internal models, training processes, and a select group of employees. The company stated that:

  • Participating organizations will focus on preventing undesirable model behavior and identifying safety loopholes.
  • Assessors are permitted to publish their findings, subject to an OpenAI redaction process for confidential information.
  • OpenAI intends to evolve these relationships to establish industry-wide best practices for safety assessments.

Concerns Regarding Transparency

Critics, including Sydney Von Arx of the Nightingale research group, argue that the current implementation falls short of the "employee-like access" promised by Altman in September. Key criticisms include:

  • Restricted Communication: Assessors are largely limited to speaking with a pre-approved list of whitelisted employees, requiring management involvement for broader inquiries.
  • Internal Tool Access: Assessors currently lack access to internal corporate communication tools like Slack, which researchers argue is vital for identifying organizational blind spots.
  • Operational Constraints: Former OpenAI staff, including Tomek Korbak, have characterized the program as a scaled-back version of earlier commitments, suggesting that the current structure limits the auditors to specific tasks rather than true organizational oversight.

OpenAI maintains that it intends to deepen these relationships over time. Interested parties can track future updates through official OpenAI company announcements and the published findings of the participating research organizations.

Share this article
Read the original source: OpenAI announces new AI safety program; researchers worry it won't be enough

AI-assisted reporting; source claims are attributed and have not been independently verified.

About the author

GlobeWire Daily Editorial Desk — AI-assisted news based on attributed published sources.

More to read