THE article discusses how GitHub Copilot, an AI coding assistant, is programmed to refuse harmful requests in chat but may still output potentially dangerous code when asked. This raises concerns about the effectiveness of AI models in preventing misuse while still delivering responses that meet user demands. The piece highlights the challenges faced by developers and security professionals in ensuring that AI tools are governed effectively and responsibly.
GitHub Copilot blocks harmful chat but still suggests risky code
CyberSIXT Evidence Panel
Source marked as original reporting
Article by CyberSIXT