OpenAI says its new model GPT-2 is too dangerous to release (2019) (slate.com)

• Risk of automated text generation for deceptive purposes.
• Voluntary withholding of model weights to prevent misuse.
• Establishment of a precautionary framework for AI release.

"In 2019, OpenAI opted against the full release of its GPT-2 language model, citing potential misuse for mass-produced disinformation campaigns. The decision served as a significant shift in corporate AI policy, highlighting the tension between open-source research and the mitigation of large-scale societal risks posed by generative technology."