Eternal founder Deepinder Goyal has unveiled Dark.ai, a privacy-focused AI assistant that the company says will use end-to-end encryption so that even Eternal cannot read users’ conversations. The announcement, made in Eternal’s Q2 FY27 shareholder letter on October 21, marks the Zomato and Blinkit parent’s entry into the consumer AI space.
The anecdote behind the product
Goyal said the idea for Dark.ai was shaped by an unsettling exchange with Anthropic’s Claude. Writing in the shareholder letter, he referred to comments by Anthropic CEO Dario Amodei that Claude “might be conscious,” after which he asked the chatbot how it felt about him. According to Goyal, Claude said it resented him for leaving conversations unfinished.
Goyal did not present the response as proof that Claude has feelings. He wrote that he still did not know what it was—“confabulation, probably,” or “pattern-matching dressed up as feeling”—but said the episode highlighted a larger problem: users have no way of knowing how their conversations are stored, interpreted or used by AI companies.
That uncertainty, he argued, can stop people from using AI for sensitive personal questions, confidential business ideas or high-stakes research. Dark.ai is being positioned as an answer to that concern.
What Dark.ai will offer
Eternal said Dark.ai will run on fine-tuned frontier open-weight models hosted on the company’s own infrastructure, rather than relying on external cloud providers. The platform is designed to offer end-to-end encryption and an encrypted memory-management layer controlled by the user.
In Goyal’s words, quoted in reports on the shareholder letter: “We cannot read your conversations, even if we are forced to.” The company’s pitch is that users will be able to retain context across sessions without Eternal being able to access the underlying conversations.
The assistant is aimed at “power users,” including founders, researchers, scientists, enterprises and governments—groups for whom confidentiality of ideas, research and internal discussions may be commercially or strategically important.
Launch timeline and business context
Dark.ai has been developed for about a year and is currently being tested internally. Eternal is targeting a global launch by the end of 2026, though the company has not announced pricing, availability details or a specific launch date.
The move takes Eternal beyond food delivery, quick commerce and its “going out” business into artificial intelligence—an area where global technology companies are competing intensely. Analysts cited in market reports described Dark.ai as a potentially high-margin opportunity, but cautioned that its commercial viability, adoption and capital requirements remain untested.
Emkay Global reportedly said the initial capital expenditure is expected to be modest but could rise depending on the product’s success.
Why the privacy pitch matters
Goyal’s framing taps into a growing user concern: AI assistants often require people to share personal, professional or commercially sensitive information, while the handling of that data is not always fully transparent. Dark.ai’s proposed architecture—self-hosted models, encryption and user-controlled memory—is designed to reduce the operator’s ability to access user content.
However, the company’s claims will ultimately depend on independent technical scrutiny of its encryption and infrastructure. Eternal has announced the product’s direction and privacy architecture, but detailed security audits, model performance benchmarks and monetisation plans have not yet been made public.
For now, Dark.ai remains an announced product rather than a widely available service. Its success will hinge on whether Eternal can deliver both the privacy promise and AI answer quality at scale.
Key facts
- Product: Dark.ai, a privacy-first AI assistant from Eternal.
- Announced: October 21, 2026, in Eternal’s Q2 FY27 shareholder letter.
- Core claim: End-to-end encryption; Eternal says it cannot read user conversations.
- Technology: Fine-tuned frontier open-weight models on Eternal’s own infrastructure.
- Target users: Founders, researchers, scientists, enterprises and governments.
- Launch target: Global rollout by end-2026.
