Former OpenAI Researcher Naomi Bashkansky Joins Conduit to Build 'Telepathy' Models

Naomi Bashkansky has resigned from OpenAI to join the startup Conduit as a Founding Researcher. The move signals a strategic shift toward brain-computer interface (BCI) technology aimed at converting human thought directly into text.

Bashkansky officially left OpenAI on July 23 and joined Conduit the following day, July 24. At the new venture, she is focusing on the development of "telepathy"—specifically, thought-to-text models trained on non-invasive neural data. According to Bashkansky, the objective is to create a seamless interface that allows thoughts to be translated into text without the need for traditional input methods, such as keyboards or voice commands.

A Shift in Research Focus

Bashkansky brings a specialized background in AI safety, interpretability, and robustness to the project. Prior to this transition, her work included affiliations with Harvard and the RAND Corporation, alongside her tenure at OpenAI. While her previous work centered on the alignment and safety of large language models (LLMs), her role at Conduit represents a pivot toward the practical application of neural data. Notably, the project emphasizes non-invasive methods, meaning the technology aims to function without the need for surgical implants in the brain.

Implications for Human-Computer Interaction

This transition reflects a broader trend within the AI industry, where high-profile researchers are leaving established labs like OpenAI to launch specialized startups. These "frontier" ventures are increasingly targeting the intersection of biology and computation to redefine how humans interact with machines. If successful, non-invasive thought-to-text capabilities could fundamentally alter the landscape of digital communication and accessibility, potentially removing the physical barriers currently associated with cognitive labor and data entry.

The Path Forward

As a Founding Researcher, Bashkansky will lead the effort to refine how neural signals are decoded into coherent language. While the core mission of building these models is confirmed, the specific technical milestones and the timeline for a consumer-ready product remain to be seen. The industry will be watching to see if Conduit can achieve high-fidelity text translation using only external sensors, a challenge that has historically plagued non-invasive BCI research.