OpenAI Teases Astra With 1979 Voice-and-Gesture Demo, but Leaves Product Details Out

OpenAI has teased its forthcoming Astra model with a 1979 MIT demonstration built around a simple idea: a computer can understand an incomplete spoken command when it also knows where a person is pointing. The archival reference signals an interest in joining voice, visual context and action, but it offers no product specifications or release date.

An old answer to an unfinished command

On September 3, OpenAI posted a two-post thread featuring MIT’s Put That There interface and credited the MIT Media Laboratory, Chris Schmandt and Eric Hulteen. The company supplied no accompanying description of Astra, two days after saying the delayed model would become available soon.

Put That There paired spoken commands with pointing gestures inside a shared visual scene. A user could point to an object, say “put that there,” then point to a destination: speech supplied the action, while gestures identified the object and place. When part of a command was unclear, the system could ask a focused follow-up question.

A broader job than the MIT demonstration

The comparison sets a demanding test. Put That There operated in a narrow graphical environment, with a constrained vocabulary and dedicated tracking hardware. Astra would need to handle messier screens, longer tasks and less predictable instructions if it is to turn the same basic idea into a general-purpose product.

The safety limits remain part of the picture

OpenAI has classified Astra as the first model to reach the Critical cybersecurity threshold in its Preparedness Framework. The company says Astra can find previously unknown vulnerabilities and develop exploits across hardened systems when given the needed tools and access.

OpenAI’s planned safeguards

• Initially restrict Astra’s most advanced cybersecurity functions.
• Use refusal training, account-level controls and monitoring against cyber misuse and unauthorized model behavior.

OpenAI delayed parts of Astra’s development while testing those stronger controls. It has also warned that the safeguards may interrupt legitimate work in ChatGPT, Codex and the API. The teaser supplies an interface ambition; a product demonstration will need to show how useful interaction and those limits coexist.