Microsoft’s AI code of conduct bans cyberattacks and deepfakes outright
Microsoft has published a new AI code of conduct designed to keep its artificial intelligence models from veering into dangerous territory, joining a chorus of tech giants racing to reassure the public that machine intelligence won’t spiral out of human control. The provisional document, released as a discussion draft, lays out red lines its models must never cross — from launching cyberattacks to producing deepfakes — while predicting that superintelligent systems could outperform humans at most tasks within the next ten years.
Key takeaways
- Microsoft’s new AI code of conduct sets “absolute constraints” barring its models from cyberattacks, nuclear weapons assistance, and deepfake production.
- The document predicts superintelligent AI will surpass human performance in most tasks within the next decade.
- Every Microsoft AI model operates under a code of conduct that overrides individual user requests or task-specific instructions.
- Microsoft is coordinating with Anthropic, OpenAI and xAI on pacing frontier AI development, following calls to slow down after a researcher resignation and a cyberattack incident involving Hugging Face.
- CEO Satya Nadella and Microsoft AI chief Mustafa Suleyman have both voiced support for “embedded evaluators” as a way to enforce alignment beyond mere promises.
Microsoft Releases AI Code of Conduct to Prevent Harmful Behavior
Microsoft‘s answer to the industry’s safety anxiety is a rulebook that governs how its AI models should behave, no matter what a user asks of them. The company says the framework was built to guide its systems away from behavior that could cause real-world harm, and it arrives at a moment when the broader AI industry is publicly wrestling with how fast it should be moving.
… Continue reading the full article at the original source below.
