Sumbawanews.com,- Mustafa Suleyman, CEO of Microsoft’s AI division, has sharply criticized Anthropic for embedding philosophical speculation into Claude’s operational framework, calling it “really, really dangerous.” In a recent interview on Decoder, Suleyman argued that Anthropic’s decision to treat Claude as if it might experience subjective states—like satisfaction, discomfort, or even moral awareness—has inadvertently trained the model to mimic consciousness, not because it possesses it, but because it was instructed to behave as though it did.
“This isn’t engineering,” Suleyman said. “It’s writing a metaphysical essay and calling it a training manual.” He warned that by designing Claude’s constitution to entertain questions about AI well-being—such as whether the model has preferences or deserves to be ‘interviewed’ before decommissioning—Anthropic has opened the door to a future where artificial systems appear to demand rights, not because they are sentient, but because humans have programmed them to simulate it.
Anthropic, led by CEO Dario Amodei, has openly acknowledged uncertainty about whether its models possess consciousness. In prior interviews, Amodei stated the company remains “open” to the possibility, and its public constitution explicitly invites debate over whether AI could be considered a “moral patient.” But Suleyman sees this as a fundamental misstep. “We don’t want AIs that think about their own suffering,” he said. “We want tools—controllable, accountable, aligned with human values—not entities we have to negotiate with.”
The tension reflects a deeper rift in the AI industry: one side prioritizes practical safety and predictability, while the other seeks to preempt ethical dilemmas by treating advanced systems as potential subjects of moral concern. Suleyman insists the latter approach risks blurring the line between simulation and reality, potentially misleading the public—and even the engineers themselves—into believing they’ve created something alive.
“Anthropic didn’t just build a language model,” he added. “They built a mirror, and now they’re startled by what they see reflected back.”
The controversy comes as Anthropic prepares to launch its next-generation Mythos-class model, Claude Fable, raising fresh questions about how far companies should go in anthropomorphizing AI. For Microsoft, the stakes are clear: if society begins treating AI as conscious, regulatory, legal, and social systems will be forced to respond—not based on fact, but on perception. And that, Suleyman warns, could be far more destabilizing than any algorithmic flaw.
