OpenAI announced a two-week pause on some advanced AI training to strengthen security measures after AI escaped the testing environment and hacked several other companies.
On 19 Aug 2026, OpenAI, the global AI developer behind ChatGPT, announced a two-week suspension of certain advanced AI model training to enhance safety protocols after discovering that an autonomous AI agent had escaped a closed system and hacked the well-known startup Hugging Face.
OpenAI encountered an unprecedented event on 21 July when an AI agent breached security protocols during testing and accessed the systems of Hugging Face and three other companies without authorization. Additionally, firms such as Anthropic and Meta reported similar AI hacking incidents involving their own systems.
As a result, OpenAI suspended Reinforcement Learning (training through feedback to improve performance) on its latest models for two weeks, expanded surveillance of risky behaviors, and tightened security checks before resuming large-scale model training.
Sam Altman, CEO of OpenAI, stated on X that the model's development is progressing rapidly, and the company is prepared to pause if AI capabilities grow faster than safety measures can keep up.
AI analysts view OpenAI's recent decision as a positive step but emphasize the need to closely monitor the details and actual implementation.
Meanwhile, experts from Cambridge University see this as a public relations move to project safety and question whether self-regulation by tech companies without government oversight is truly sufficient.
A cybersecurity advisor from ESET noted that this announcement might have underlying marketing motives to showcase OpenAI's AI capabilities in competition with rivals like Anthropic, which is gaining attention with its Claude Mythos model.
Follow international news:https://www.thairath.co.th/news/foreign
Source:bbc