Mostik explores a way for AI models to communicate more directly

In short: Wired reports that a Russian startup called Mostik is developing a method for AI models to communicate in a more direct way, which the writer compares to “machine telepathy.”

What happened

Wired senior writer Will Knight described meeting Russian mathematicians connected to a startup called Mostik. They showed him a technique they say can let AI models pass information to each other in a way that is not like normal back and forth text.

Today, when people connect different AI systems, it usually looks like messaging. One system produces text, another system reads that text, and then responds. The idea Mostik is exploring sounds more like sharing meaning directly, like handing someone a sealed note instead of reading it out loud (an analogy for passing internal signals rather than only words).

The Wired story frames this as an early look at a new approach to combining the strengths of different AI models. In simple terms, it is about getting separate AI systems to cooperate more smoothly, so they can solve tasks together instead of acting like strangers on a group chat.

Why it matters

If AI models can share information more efficiently, it could make AI assistants faster and more consistent when they use multiple tools behind the scenes. It also raises fresh questions about safety and oversight, because it can be harder to understand and monitor what systems are “saying” to each other when that communication is not plain language.

Source: Wired