Pandorex
AI & Chips

Apollo moves closer to OpenAI’s development process

Published Updated Pandorex Redaktion·1 min read
—

Keeping AI under control is a question Apollo Research plans to examine closer to OpenAI’s development process. On November 16, it announced a four-month pilot involving six researchers. They will investigate whether models undermine their own or their successors’ control mechanisms during training, testing and internal use.

The project forms part of a broader program OpenAI announced three days earlier with Apollo, METR, Transluce and Goodfire. The organizations will examine different questions, ranging from misbehavior during training to the inner workings of models. That takes the assessment beyond the behavior of a finished product.

Apollo plans two public reports. Its agreed access remains limited, however, and OpenAI can withhold information, including to protect intellectual property. Apollo may publicly explain these restrictions and says the pilot does not yet meet all its criteria for fully embedded evaluations.

The reports should reveal what closer access makes possible. For now, the announcement opens an assessment process; a safety verdict on the models is still to come.

Sources and references

Sources used for the facts and context in this article.

  1. Apollo Research, 16.11.2026apolloresearch.ai
  2. OpenAI, 13.11.2026openai.com

How Pandorex researches and corrects articles

Comments

Sign in to write a comment.

Swipe up
Next Article

TSMC Hits Record Third-Quarter Revenue, but the Margin Picture Is Still Missing

AI & Chips