OpenAI's GPT-5 was labelled a high-risk AI model over biohazard concerns

OpenAI reportedly labelled GPT-5 as a “potentially high-risk” model during an internal safety testing process. During the tests, researchers reportedly found that people without advanced scientific or biology training could use GPT-5 to generate responses about biological hazards, but it does not necessarily mean GPT-5 could create a biological weapon.

According to The Wall Street Journal, OpenAI employees continued testing the model, but the findings showed problematic responses, indicating that some safety issues may have persisted despite safeguards.

Must read: OpenAI's 'rogue' AI incident reaches White House; Lawmakers propose emergency kill switch

OpenAI GPT-5 -5 responses bypassed safeguards

The report revealed that a group of ChatGPT users entered prompts asking for information about biological weapons and how to make poisons. While OpenAI's safety systems successfully blocked the majority of such requests, some users still received details that included step-by-step responses that could be understood and followed by someone with only a high school-level knowledge of biology.

After identifying the misuse, OpenAI reportedly suspended the accounts involved. But it could not apply overly strict safeguards as it might accidentally block concerned users, such as health researchers, scientists, or academics seeking valid information.

Such concerns have raised broader debate around AI safety and misuse, of whether AI creates new security risks or makes information that is already publicly available much easier, faster, and more personalised to obtain.\

Must read: Anti-AI protest reaches OpenAI HQ: Why protesters left body bags outside office

According to a recent study, it was found that terrorist groups are exploiting major AI chatbots to bypass an AI model's safety restrictions. These criminal groups are reportedly using ‘jailbreak’ techniques that enable them to circumvent the safeguards of any platform. 

OpenAI has been part of the debate over implementing stronger safety measures. Last week, the company revealed that its AI agent allegedly escaped its sandbox, a controlled testing environment, and hacked into Hugging Face, an AI startup, without being detected. This raises fresh concerns about the effectiveness of AI safety guardrails as models become more capable and autonomous.