Naomi Bashkansky, a former employee at OpenAI, has taken on the role of founding researcher at Conduit, a startup focused on training AI models for "non-invasive reading of human thoughts."

Two weeks ago, I resigned from OpenAI to join Conduit as a founding researcher, where we're training models to non-invasively read the human mind.

I've written some thoughts about what telepathy could look like by 2035 and how to get there: pic.twitter.com/cat0d4My15

— Naomi Bashkansky (@NaomiBashkansky) August 5, 2026

In her blog, Bashkansky expressed her belief that in the coming years, thoughts will become the primary method for interacting with artificial intelligence. She emphasized that this approach should empower humans rather than replace them.

"These will not be super-intelligent automated interns or colleagues that you are desperately trying to keep up with. It will be a natural, joyful extension of yourself," she stated.

Bashkansky predicts that by 2027, individuals will be able to transmit thoughts to their agents via Bluetooth, which will interpret them into prompts and actions.

A theoretical example of translating thoughts into AI. Source: Bashkansky's blog.

She forecasts that by around 2030, the industry will broaden and begin utilizing Conduit's telepathic technologies. By 2035, she believes AI will no longer be viewed as "something alien" by people.

She referenced industry leaders Elon Musk and Sam Altman, who predict an increasing convergence between the human brain and artificial intelligence.

"The degree to which our AIs will be superhuman in 2035 directly depends on how superhuman we, as a society, choose to make them. We might opt for a slower pace of development. But I love that this is literally a vision of the future where humanity is involved in the process, and AI directly enhances human capabilities rather than replacing us," she remarked.

Bashkansky noted that Conduit aims to train models based on brain activity:

"In theory, it’s simple. The input is brain activity, and the output is what the person was doing at that moment, for example, what text they wrote. Based on brain activity, we want to predict an outcome semantically similar to what the person wrote."

Conduit is based in San Francisco and is developing non-invasive neurointerfaces. The company has not disclosed any information regarding its funding or valuation.

Some users commented on Bashkansky's post, criticizing the term "non-invasive." One commenter pointed out that the concept of reading thoughts is difficult to implement without direct access to the brain.

It is worth noting that in June, Meta introduced a system that reconstructs typed text from brain activity recordings without surgical intervention, achieving an average word accuracy of 61%, compared to approximately 8% from previous methods.