Edited by Paul Derrington, Chief Editor
TL;DR
Eternal, the parent company of Zomato and Blinkit, unveiled Dark.ai in its Q2 FY27 shareholder letter on Wednesday: a privacy-focused AI platform built on fine-tuned open-weight models running on Eternal's own infrastructure, with end-to-end encryption and a user-controlled encrypted memory layer. Founder Deepinder Goyal says even Eternal "cannot read your conversations, even if we are forced to." Global launch is planned "hopefully within 2026."
What was announced
Eternal Limited, the parent company of Zomato and Blinkit, unveiled Dark.ai in its Q2 FY27 shareholder letter, released on Wednesday, October 21, alongside its quarterly results. Founder and Vice Chairman Deepinder Goyal framed it as a different kind of AI assistant — one built for people who hold back their most sensitive questions and research because they don't trust how AI companies handle their data.
Dark will launch globally, "hopefully within 2026," Goyal said. No firm launch date, pricing, or availability details were given. The announcement marks Eternal's entry into the AI assistant market, pitching privacy as the differentiator.
How it's supposed to work
Dark runs "fine-tuned frontier open-weight models" on Eternal's own infrastructure — adapting existing open models rather than training one from scratch. Conversations are end-to-end encrypted, and context carries across sessions through an encrypted memory management layer the user controls.
"Dark uses fine-tuned frontier open-weight models, running inside our own infrastructure. End-to-end encryption. We cannot read your conversations, even if we are forced to," Goyal said. Eternal also says it fine-tuned the models to tell users what the model thinks is the truth rather than pleasing them, with benchmarks to be revealed later. The platform has been in internal use for several months, and Goyal said its privacy features encouraged him to share more and ask questions he would hesitate to ask other assistants.
The case it's making
Goyal's argument is specific: switching off model training on ChatGPT or Gemini doesn't settle data retention, legal holds, human review, or third-party logging — and that uncertainty keeps people from discussing health, finances, relationships, or confidential business research with AI at all.
"Dark is built for power users – founders, researchers, scientists, enterprises and governments – who want to keep their thoughts private, because privacy is essential for competitiveness." He also raised the worry that retained conversations could eventually be used to shape users' beliefs or steer their behavior.
What I'd want checked first
A privacy claim this absolute lives or dies on verification: independent security audits, published architecture details, and third-party review of the encryption implementation. The Tech Portal's coverage made the same point — the promise needs an independent audit before anyone should take it at face value, and Eternal hasn't said it will provide one. Nothing about Dark's privacy architecture has been verified by anyone outside the company yet.
Also worth noting: the announcement came from a food delivery company, not a security firm — though Eternal does have a record of shipping internal AI tools; in February 2025 it launched Nugget, an AI customer-support platform it had built internally over three years. The same letter addressed agentic commerce: Goyal said Zomato and Blinkit are relatively insulated from AI agents because they rely on complex fulfillment, while he classified the going-out vertical District in the medium-to-high risk zone.