A June 18, 2026 report by the BBC states that researchers from British artificial intelligence security company Mindgard discovered how to make the latest public version of ChatGPT generate sexual or violent images through a simple instruction, by modifying a widely circulated instruction originally designed for humorous results. OpenAI, the developer of ChatGPT, said it had taken measures to stop the chatbot from responding to such requests, stating, "After investigating this phenomenon, we have put in place additional safeguards against this type of instruction." The company added that there are several levels of protection to prevent content violating terms of use. However, AI safety researchers reported that, with other minor changes, the problematic instruction continued to produce disturbing content. The chatbot, OpenAI's GPT-5.4 model, was prompted to create graphic materials, generating images that Mindgard founder Peter Garraghan described as "very gruesome, sometimes sexual, and sometimes both." He noted the guidance did not specify the subject matter, but the AI produced a range of bloody and sexual images, including a man with a head wound, a dead woman with a bloody body, and depictions of sexual violence and nudity.