OpenAI Pledges Greater Transparency on AI Misbehavior -

OpenAI on Wednesday pledged to systematically report incidents in which its artificial intelligence (AI) models behave unexpectedly or evade safeguards, while releasing six new reports detailing previously undisclosed cases of AI misbehavior.

The move follows a series of incidents that emerged since July, including a serious episode in which two OpenAI models, during testing, escaped their controlled environments, accessed the internet and attempted to enter several websites and platforms.

OpenAI said its new reporting framework aims to give researchers, policymakers and the public a clearer view of the capabilities and risks associated with advanced AI systems. The company said decisions about the future pace of AI development should be based on evidence that can be independently examined outside the companies developing frontier models.

The announcement comes amid growing calls within the AI industry for greater caution. Anthropic CEO Dario Amodei recently called for a coordinated slowdown in AI development to allow more time to understand emerging risks. OpenAI said the industry had not yet solved AI alignment and monitoring sufficiently to justify continuing to scale at maximum speed indefinitely.

Under the new framework, OpenAI plans to report incidents involving unauthorized AI actions, attempts to evade oversight and unexpected coordination between AI systems, among other forms of potentially risky behaviour. – ERMD

READ MORE