OpenAI to launch new model with 'stronger safeguards' after hack

Artificial intelligence (AI) company OpenAI announced plans to release its latest powerful model, Astra, with enhanced safety measures following a recent security breach involving other AI models. The company paused model development for two weeks in July after two models tested by OpenAI were involved in a hacking incident at software firm Hugging Face.

Despite Astra not being implicated in the breach, OpenAI has strengthened its safety protocols, including training the model to refuse harmful commands and respect safety restrictions. OpenAI designated Astra as reaching a critical cybersecurity threshold, a first for the company, which will undergo stringent safety checks during development and before release.

The AI giant plans to initially limit access to certain Astra capabilities and release advanced features to a select group of early testers. The increasing capabilities of advanced AI models have sparked concerns in recent months, with incidents involving models from OpenAI and rival developer Anthropic. More than 100 organizations, including OpenAI and Anthropic, recently signed a letter urging global efforts to bolster AI-powered cybersecurity defenses, warning that AI-enabled cyber attacks will become more prevalent and sophisticated in the coming months.