Microsoft’s AI boss Mustafa Suleyman has thrown a wrench into the AI safety debate, suggesting that the concept of alignment may be flawed and that companies like Anthropic are making it worse. In a detailed interview, Suleyman discusses the necessity of containment and the potential dangers of unguarded AI models.

‘If I designed a car and 10 percent of the time the brake pedal decided to go attack my neighbor’s house, I would be like, “This car doesn’t work. The very technology of brakes is broken. I need a new idea,”’ Suleyman argues, drawing a stark analogy between AI and automotive safety.

Microsoft’s Humanist AI Code of Conduct, a 37-page document, outlines the company’s principles and philosophy on AI development. Suleyman believes that the industry must focus on containment and alignment to ensure that AI models are controllable and aligned with human values.

The debate around AI safety and regulation is gaining traction, with concerns over consciousness and the potential for hacking. Suleyman’s critique of Anthropic’s approach highlights the need for a more cautious and controlled approach in AI development.

‘We need to slow down before we kill us all,’ Suleyman warns, emphasizing the importance of safety guardrails in AI systems.