Researchers at OpenAI studied the potential risks of releasing an open-weight version of their GPT-OSS large language model. They introduced a method called 'malicious fine-tuning' (MFT) to test how far they could push the model's capabilities in two areas: biology and cybersecurity. The goal was to estimate the worst-case scenario for such a release. Their findings suggest that an open-weight LLM like GPT-OSS could be used for malicious purposes, highlighting concerns about the potential risks of releasing powerful AI models without proper safeguards. --- Why it matters: This research is important because it highlights the need for careful consideration when releasing powerful AI models to the public. If not properly secured, such models could be exploited for malicious purposes, which has significant implications for fields like cybersecurity and biology. Source: https://openai.com/index/estimating-worst-case-frontier-risks-of-open-weight-llms