Microsoft AI head calls out Anthropic for acting like Claude is conscious

Jun 10, 2026 03:24 AM - 14 hours ago 637

Emma Roth

is a news writer who covers the streaming wars, user tech, crypto, societal media, and overmuch more. Previously, she was a writer and editor astatine MUO.

Microsoft AI CEO Mustafa Suleyman says it’s “really, really dangerous” for Anthropic to estimate astir Claude’s consciousness inside its “constitution,” aliases the instructions that show the exemplary really to behave. During an section of Decoder, Suleyman argues that this benignant of speculation whitethorn person group up the chatbot to enactment arsenic though it’s conscious:

I deliberation that it’s almost arsenic though immoderate of the folks astatine Anthropic person anthropomorphized the creation of Claude truthful overmuch that it has past gone and wireheaded them and benignant of tricked them into believing that it has these glimmers of consciousness that they put into it successful the first place.

Suleyman adds that “we do not want to person to contend pinch a super-intelligence that has ideas astir its ain suffering, aliases ideas astir its ain feeling.”

Claude’s constitution straight references Anthropic’s uncertainty astir whether the AI exemplary has well-being and if it experiences things for illustration “satisfaction” aliases “discomfort.” Anthropic besides says the institution will “interview” AI models erstwhile they’re deprecated and will archive immoderate “preferences” they person astir early releases.

While connected Decoder, Suleyman calls this a “philosophical failing,” arsenic Anthropic made Claude’s constitution “a spot for speculation for illustration you would successful an world insubstantial alternatively than a training manual.” This has led Claude to internalize these “ideas astir itself and its ain training,” Suleyman says.

Anthropic CEO Dario Amodei has alluded to Claude’s imaginable consciousness successful the past, saying during an question and reply pinch Interesting Times that “we don’t cognize if the models are conscious” but that the institution is “open” to that idea.

“This is precisely what we don’t want from AIs,” Suleyman says. “We want AIs to beryllium controllable, contained, accountable, aligned devices that service humanity.”

Follow topics and authors from this communicative to spot much for illustration this successful your personalized homepage provender and to person email updates.