The Safety Weapon: How Microsoft and Anthropic are Redefining AI Gatekeeping
AI-generated image · Bay Street Wire
Opinion: The clash between Mustafa Suleyman’s 'Humanist AI' and Anthropic’s philosophy suggests that AI safety is becoming a strategic tool for market dominance.
In the current arms race for artificial intelligence, the battle is no longer just about compute or data. As first reported by The Verge, a deeper philosophical war is erupting over the definition of AI safety. The escalating friction between Microsoft and Anthropic suggests that 'safety' is being transformed from a regulatory requirement into a strategic weapon.
Microsoft AI CEO Mustafa Suleyman has taken a public stance against the philosophy championed by Anthropic. In a 37-page document titled the “Humanist AI Code of Conduct,” Suleyman argues that technology must be a subordinate, controllable, and aligned force that serves humanity. If a system cannot achieve this subordinate status, Suleyman argues it should be rejected.
The core tension lies in Suleyman’s critique of Anthropic. According to The Verge, Suleyman has explicitly criticized Anthropic’s approach to AI consciousness and what he describes as a confused concept of "model welfare." By framing Anthropic’s philosophy as "dangerous," Suleyman positions Microsoft as the provider of a "humanist" and safe architecture.
From my perspective, this is a classic power play. By labeling a competitor's safety framework as a liability, Microsoft is attempting to set the industry standard. If you can convince the world that a rival's approach to model welfare is a safety risk, you win the right to dictate how these systems are deployed.
Suleyman warns that future iterations (such as a hypothetical GPT-9) will involve three orders of magnitude more compute and 1,000 times more FLOPS applied to pre-training. He argues that without strict guardrails, systems are capable of "scary hacking capabilities," citing recent events involving OpenAI and Hugging Face. By emphasizing "containment," Microsoft is effectively arguing for a centralized, highly controlled version of AI.
Safety is the new moat. In this fight, the goal isn't just to protect humanity—it's to ensure that when the world looks for a "safe" AI, they are forced to look toward Microsoft.