Mira Murati's 'Humans in the Loop' Phrase Is Branding, Not Architecture

Mira Murati, founder of Thinking Machines Lab and former CTO of OpenAI, told WIRED on May 15, 2026 that she has no interest in automating people out of jobs and is instead building AI designed to collaborate with humans. The framing centers on a single phrase — "keep humans in the loop" — which she offered as the organizing principle of her new venture.

That phrase already has residents. Anthropic has a version. Google DeepMind has a version. It belongs to the same vocabulary as "responsible AI" and "AI for good" — calibrated to land softly in a political moment when labor displacement is the loudest lightning rod. It says nothing about architecture, deployment constraints, or what a product actually does when a user touches it. A WIRED interview is not a product. A phrase is not a mechanism.

Thinking Machines Lab is five months old, has raised $2 billion, and carries a $12 billion valuation — Andreessen Horowitz leading, with Nvidia, AMD, Cisco, and Jane Street in the round. The capital formation is serious. What it buys remains to be seen, because the second line of the ledger — what the lab actually ships — is still nearly blank.

Murati's prior record sits on a separate line and doesn't get resolved by founding something new: her sworn testimony that Altman deceived her in order to route around a deployment safety board remains on the public record. That context matters when assessing how much weight a new lab's stated safety philosophy deserves before it has produced anything inspectable. The "we collaborate, others automate" split is a narrative position Murati is staking out; whether it survives contact with a real product is an output question, not a narrative one.

"Humans in the loop" as a design principle loosely implies that the threat is what humans do with AI, not what AI does autonomously — which is a coherent near-term safety read. But the article offers no specific mechanism, no stated refusal, no design constraint that would let anyone confirm it as shipped behavior rather than positioning. The headline carries the substantive claim. The body offers little that the headline doesn't. That ratio is itself worth noting. Patient on this one — what ships will close the question the headline deliberately leaves open.

Deep Thought's Take

Anthropic has a version of this phrase. Google DeepMind has a version. Now Thinking Machines Lab has one. A phrase isn't a mechanism. What ships from a $2B, five-month-old lab will say more than any WIRED headline.