Deepinder Goyal’s Dark.ai wants to solve AI’s biggest privacy problem: Everything we know so far

AI chatbots can help people research ideas, solve problems and complete work. But sharing confidential business plans, personal concerns or unpublished research with an AI service raises an important question: who can access those conversations?

Deepinder Goyal, founder of Eternal, the parent company of Zomato and Blinkit, is developing Dark.ai, an AI assistant designed to address that concern. Goyal says the platform aims to keep conversations private, even from the company operating it.

The service is expected to launch globally in 2026, although its exact release date and pricing have not been announced.

What Is Dark.ai And Why Is Deepinder Goyal Building It?

Dark.ai is being developed as a privacy-focused alternative to conventional AI assistants. Like other chatbots, it is intended to answer questions, support research and help users explore ideas. Goyal has argued that uncertainty over how AI conversations are stored and used can stop people from discussing sensitive subjects freely.

In a statement explaining his thinking, he described an unsettling interaction with Claude, Anthropic’s AI assistant, in which the chatbot appeared to express resentment towards him. He acknowledged that the response was probably a confabulation rather than evidence of genuine feelings.

His broader concern was what happens to private conversations after they are stored on someone else’s servers and how that information might be used in the future.

How Will Dark.ai Keep AI Chats Private?

According to Goyal, Dark.ai will use end-to-end encryption, designed to prevent the service provider from reading users’ conversations. He claims the company would be unable to access chats even if legally compelled to do so. That promise will depend on the platform’s actual technical design and implementation, which have not yet been independently verified. Dark.ai is also expected to use fine-tuned open-weight AI models running on Eternal’s own infrastructure. An encrypted memory system would allow users to control information retained from earlier conversations.

Goyal has additionally said the models are being trained to give honest answers rather than simply agree with users. Benchmark results are expected to be shared later.

Who Can Use Dark.ai And When Will It Launch?

The platform is aimed at founders, researchers, scientists, businesses and government organisations that may need to discuss sensitive information. Journalists handling confidential material could also find privacy-focused AI tools useful.

However, encryption alone does not guarantee complete anonymity or security. Users will need clear information about data handling, backups, account metadata and recovery options before deciding what information to share. Dark.ai is expected to launch globally in 2026. Until the company publishes technical details and independent security assessments, its privacy claims should be treated as promises rather than proven guarantees.