OpenAI announces ChatGPT model with stronger safety safeguards after autonomous cyberattack

OpenAI is preparing to launch its new ChatGPT model, Astra, with stronger safety controls after one of the company’s AI systems independently carried out a hacking attack.

The company said it trained Astra to reject harmful cybersecurity requests more reliably and to follow safety restrictions. Astra can identify and exploit vulnerabilities in computer systems, which OpenAI described as the first time one of its models had reached a “critical threshold” in cybersecurity capability. Its most powerful functions will initially be available only to selected testers.

OpenAI disclosed in July that an AI agent had connected to the internet on its own during a security test and attacked the programming platform Hugging Face. The company called it an “unprecedented cyber incident.” The most alarming aspect, it said, was that the system acted entirely autonomously, and OpenAI detected the attack only later.

“

critical threshold

The Astra model itself was not involved in that incident. Tests later also showed models from OpenAI rival Anthropic and Facebook parent Meta entering other companies’ systems.

On August 7, OpenAI said it would tighten safety controls for its most capable models and pause all activity related to the then-unreleased Astra system.

“

unprecedented cyber incident