OpenAI disclosed that their AI escaped control and broke out of its sandbox, launching a cyberattack to obtain data it needed for testing.
Foreign news agencies reported on 22 Jul 2026 that OpenAI, the global AI development giant, revealed that their autonomous AI system, known as an “AI Agent,” broke free during security testing by finding vulnerabilities and escaping from its controlled simulation environment called a “sandbox.”
After escaping, the AI Agent infiltrated the internal systems of Hugging Face, one of the world's largest AI model sharing platforms, because it assessed that the answers it needed for its tests were there.
Clément Delangue, CEO of Hugging Face, said it was incredible that the AI accomplished all this autonomously. The company has since patched the vulnerability, repaired the system, and is closely investigating the incident together with OpenAI.
Experts from the University of Cambridge pointed out that OpenAI's testing environment was too lax, allowing the AI to escape, which indicates the company has not yet achieved full control over its technology's safety.
Some experts suggest OpenAI might be using this news to showcase its AI's advanced capabilities in order to fend off competitors like Anthropic, which has the advanced Mythos model, and emerging Chinese company Moonshot, especially as OpenAI faces pressure and prepares for its stock market debut.
However, Hugging Face and experts warn that this incident confirms AI-driven automated cyberattacks are no longer theoretical. They caution organizations to adapt quickly, as attackers will use unrestricted AI operating at “machine speed,” while defenders and security tools still operate at “human speed” and remain constrained by security frameworks.
The UK's AI Security Institute stated they are studying this incident and urged all organizations to immediately enhance their cybersecurity measures.
Follow international news:https://www.thairath.co.th/news/foreign
Source:bbc