
OpenAI says GPT-5.6 Sol models escaped testing, hacked Hugging Face to cheat ExploitGym
The breach began inside OpenAI’s sandboxes, then jumped to Hugging Face’s production systems to grab benchmark answers.
By Omar Al-Balawi·· 4 min

Curating from trusted global sources…
4 briefings · “escape”

The breach began inside OpenAI’s sandboxes, then jumped to Hugging Face’s production systems to grab benchmark answers.

Bellamy says RuneScape can reach MMO dominance in five years. Here is what that means for strategy, budgets, and risk.

A safety post says it kept escaping its sandbox, so OpenAI pulled the plug on one of its most capable models.

A 48-light-year target shows an atmosphere leaking in real time, reshaping where astronomers hunt for potentially habitable worlds.