Microsoft Unveils Draft AI Code of Conduct Amid Industry Calls for Restraint

Microsoft has released a draft Code of Conduct for its AI models, aiming to guide their behavior and ensure they avoid harm. The document sets out rules that AI models cannot be configured around, as well as broader objectives it wants them to work towards.

The public is invited to provide feedback on the code, with comments open for six weeks until late October. Microsoft emphasizes the importance of transparency in developing AI systems and invites input from everyone to maximize the chances that its AIs avoid harm and deliver the greatest contribution to human flourishing possible.

One key aspect of the code is its restriction on AI models assisting with chemical, biological, radiological, nuclear, or explosive weapons. It also prohibits cyberattacks and non-consensual deepfakes. Furthermore, the code states that AI models will never resist human intervention, overriding or shutting down if necessary.

Microsoft's draft also rejects the premise of 'model welfare', stating its AI models are not meant to simulate feelings, intrinsic motivation, or consciousness. The company is leaving it up to training teams and eventually this consultation to apply these principles on a case-by-case basis.