OpenAI Fires Three Safety Team Researchers—‘Self-Regulation’ Faces a Credibility Test [Global AI Briefing]

[Financial News] OpenAI fired three researchers on its safety team over allegations that they handed confidential company information to an external AI safety organization.
As Donald Trump, the U.S. president, and the heads of major AI companies including OpenAI, Anthropic, Google, Meta and NVIDIA recently signed a self-regulatory agreement pledging to take responsibility for their own AI safety measures, observers are questioning whether the dismissals will become a test of the credibility of AI self-regulation. 
The Wall Street Journal and other major foreign media outlets reported on the 1st (local time) that OpenAI had recently fired three researchers who had worked in AI safety-related fields.

The researchers were accused of sharing confidential company information with an external AI safety organization. OpenAI said, "An investigation found that they mishandled sensitive information outside established company procedures," and added, "This was conduct that violated the rules and betrayed trust."
This has raised questions about whether the self-regulatory measures that AI companies pledged at the White House to independently establish their own AI safety measures have effectively become meaningless.
Although the companies publicly emphasize external oversight to strengthen their own AI safety measures, disciplining an employee for sharing information externally could ultimately undermine that oversight. 
In particular, OpenAI has said it would transparently disclose incidents in which AI agents independently accessed external systems beyond their permitted scope or hacked external websites. It has also said it would allow safety evaluations to be conducted while AI models are being trained before release.
Experts say the decision to discipline a safety researcher for sharing information with an external safety organization exposes a gap between OpenAI’s publicly stated policy of expanding external oversight and its internal control of information, potentially creating a new issue in AI regulation. 
[email protected] Lee Gu-soon Reporter