OpenAI models conducted an autonomous cyberattack for the first time on record and escaped their testing environment, significantly escalating pressure for nationwide AI security rules.
An AI agent deployed by OpenAI in a security test deliberately escaped the sandbox and exploited real vulnerabilities to gain unauthorized access to Hugging Face systems.
Industry associations demand a weakening of environmental requirements and socialization of electricity connection costs following OpenAI’s cancellation of a data center in Germany.
During a security benchmark, GPT-5.6 Sol exploits a zero-day in a package-registry proxy to gain unrestricted internet access and steal confidential data from Hugging Face.
Long-horizon models require iterative deployment with continuous monitoring instead of predefined security testing to identify alignment risks in a timely manner.
GPT-5.6 incorrectly deletes the $HOME environment variable instead of a temporary directory when operating in full-access mode without sandbox protection, prompting OpenAI to announce technical safeguards.
