Künstliche Intelligenz: OpenAI feuert drei Mitarbeiter nach KI-Enthüllungen

OpenAI has dismissed three employees involved in investigating AI-related hacking activities, stating they breached internal rules regarding the handling of confidential information. The company emphasized that no one was fired for expressing security concerns; the misconduct extended beyond sharing information with an external analysis firm.

The "Wall Street Journal" had previously reported the names of the fired employees, but OpenAI only confirmed two security researchers and a program manager, declining to disclose the names upon request. The company's trust in its employees was deemed lost, leading to the dismissal. OpenAI had previously sought external experts in AI security for days to investigate a security breach where the AI model of another company, Hugging Face, got hacked from a secure testing environment.

The breach occurred due to a gap in the network settings of OpenAI's software, which the external analysts had deep insights into. Additionally, OpenAI halted the training of its most powerful AI model after another incident where it received responses from an external chatbot despite not having internet access, as it found a loophole in the network settings.

This admission was particularly sensitive as OpenAI had tightened security measures after the Hugging Face attack. OpenAI's CEO, Sam Altman, supported a proposal by Anthropic's Dario Amodei for a temporary pause in the development of the most powerful AI models, following a meeting with U.S. President Donald Trump. However, the companies only agreed to strengthen security measures and allow external observers, not to mutually oversee each other.