Sam Altman has confirmed what OpenAI's job listings have long hinted at: the company behind ChatGPT is building a humanoid robot.
Asked on the Sources podcast whether OpenAI will actually build one, the CEO's answer was brief. "We will." For more context on this story, see our ongoing AI news.
Why a human shape
For Altman, the rationale for a two-legged, two-armed machine is practical rather than aesthetic. "The world is very much designed for people," he said on the podcast, pointing out that doors, keyboards, and tools are all built around a human body. A robot that needs to work in human spaces should therefore match the human form factor. "Matching that form factor seems good," he said.
The argument echoes the one made by virtually every humanoid developer: homes, factories, hospitals, and offices were not designed for wheels and tracks, so the most general-purpose laborer is one shaped like the people the environment was built for.
The brain is the hard part
What the robot will actually look like is, in Altman's telling, a secondary question. Those design details are "less important than really figuring out ... the brain that makes the robot work," he said.
The distinction matters because the failure points of humanoid robotics are mostly cognitive. A capable machine has to interpret what its cameras and sensors see, understand an instruction, plan a sequence of movements, execute them with the right amount of force, and adapt when the physical world refuses to cooperate. Picking up a mug sounds trivial until the grip slips, the handle faces the wrong way, or the mug is a paper cup that crumples under too much pressure. As Engadget's analysis of the announcement put it, giving a machine two arms and two legs accomplishes little if it does not know what to do with them.
OpenAI's hiring activity shows where it thinks the difficulty lies. The company's robotics roles span motion control, state estimation, actuator control, and whole-robot behavior, along with the machine-learning infrastructure needed to train physical AI systems.
OpenAI's robotics history
This is not OpenAI's first brush with robots. Its Dactyl research project trained a robotic hand to manipulate real objects, with all of the training done in simulation before the skills were transferred to hardware. The team later applied the same approach to solving a Rubik's Cube one-handed, an early demonstration of the sim-to-real problem that still defines the field: a model can learn what should happen in software, but physics gets the final vote.
The company's current job postings suggest it is taking hardware more seriously than the "brain over body" framing might imply. Listings indicate work on custom electromechanical actuators — the motors, transmissions, sensing, and thermal systems that make joints move — as well as manufacturing and supply chain positions aimed at eventually producing hardware at scale.
Timeline: a demo in 2027
Anyone hoping to see an OpenAI robot soon should temper expectations. Altman told Fortune in September: "I think we'll have a cool demo in 2027 ... I don't think there will be robots walking the streets for a few years longer."
OpenAI has not disclosed what the demo will showcase, what its humanoid will look like, or when anyone outside the company will get near one. The vagueness is standard for the category, where staged videos routinely outpace real-world capability, and where "autonomous" demonstrations sometimes rely on teleoperation or carefully staged conditions.
A crowded field
OpenAI is arriving late to a race that is already well underway. Tesla has spent years developing Optimus, its general-purpose humanoid, with Elon Musk repeatedly promising commercial sales that have yet to materialize at scale. China's XPENG grabbed headlines recently when its IRON humanoid walked autonomously off a production line, and a wave of venture-backed startups — from Figure to Apptronik to 1X — are chasing the same goal of general-purpose labor.
What OpenAI brings to the contest is the piece Altman says matters most: frontier AI models. If the company's reasoning systems can be adapted to control physical machines reliably, the intelligence layer that other developers struggle with could be OpenAI's core advantage. That is presumably the bet, and it explains why the company is comfortable talking about form factors as a solved-adjacent problem while it staffs up on perception, control, and actuation engineers.
What to watch
The concrete signals to watch are OpenAI's robotics job postings, which reveal the engineering roadmap; any sim-to-real research the company publishes, which would show how its models handle physical reasoning; and whether the promised 2027 demo involves genuinely autonomous behavior in unstructured environments.
For an industry that has spent two years arguing about whether humanoids are a real product category or an investor story, OpenAI's entry is a meaningful vote of confidence. The company that put chatbots on every desk now wants to put a robot in every workplace — and it is telling everyone in advance that the software, not the hardware, is where the battle will be won.
---
Stay Ahead of AIGet the latest AI news, analysis, and breakthroughs — all in one place.
Read more AI news →
