3 ms·
Do Users Write More Insecure Code with AI Assistants?
- pmontra 4y agoContrary to the law of headlines the answer is yes. > Overall, we find that participants who had access to an AI assistant based on OpenAI's codex-davinci-002 model wrote significantly less secure code than those without access.
- dtagames 4y agoIf this means that people who forget to pay attention to security are likely to use code generation, that makes sense. But don't blame the code generator. If you are prompting an AI for your security model, you're already in a bad place. A vast amount of code that's not directly related to security, however, can be usefully generated when using AI tools interactively, line by line.