The British Institute for Artificial Intelligence Security published an article on the 12th saying that with the rapid improvement of cutting-edge artificial intelligence capabilities, security risks are moving forward to the model development stage, and existing safety assessment methods are difficult to keep up with technological development, so it is necessary to explore new technical security practices to ensure the safe implementation of cutting-edge AI development activities. The article points out that in the past, AI model security assessments mainly focused on ensuring security during public release, but now the autonomy of some AI systems has been enhanced, especially in the field of cybersecurity. Even if the model has not yet been publicly released, it may pose risks during training, internal deployment, and evaluation.

The British Institute for Artificial Intelligence Security published an article on the 12th saying that with the rapid improvement of cutting-edge artificial intelligence capabilities, security risks are moving forward to the model development stage, and existing safety assessment methods are difficult to keep up with technological development, so it is necessary to explore new technical security practices to ensure the safe implementation of cutting-edge AI development activities. The article points out that in the past, AI model security assessments mainly focused on ensuring security during public release, but now the autonomy of some AI systems has been enhanced, especially in the field of cybersecurity. Even if the model has not yet been publicly released, it may pose risks during training, internal deployment, and evaluation.