loading
loading
Zomato and Blinkit parent Eternal has unveiled Dark, a privacy-focused AI platform that founder Deepinder Goyal hopes to launch globally within 2026. In Eternal’s Q2 FY27 shareholder letter, Goyal said the platform is aimed at power users, including founders, researchers, scientists, enterprises, and governments, who want to keep their conversations and research private. Dark has been in internal use for a few months, he said, adding that its privacy features had encouraged him to share more information and explore questions he would hesitate to ask other AI assistants. The platform uses fine-tuned frontier open-weight models running on Eternal’s own infrastructure. Unlike building a model from scratch, this approach involves adapting existing models whose trained parameters are available for developers to use and modify. According to Goyal, Dark uses end-to-end encryption, alongside an encrypted memory management layer controlled by users, to retain context across sessions. “We cannot read your conversations, even if we are forced to,” he said. He positioned Dark around concerns that users’ interactions with mainstream AI assistants could be retained or accessed by others, discouraging them from sharing sensitive information. He argued that disabling the use of conversations for model training does not, by itself, address questions around data retention, legal holds, human reviews, or third-party logging. Eternal's AI Ambitions Goyal said concerns around data retention, human reviews, legal holds and third-party logging on mainstream AI platforms could limit the extent to which people use AI for sensitive or commercially valuable work. He argued that founders, researchers and scientists may hesitate to share breakthrough ideas with AI tools over fears that their information could be accessed by others. Similarly, individuals may avoid asking AI assistants sensitive questions about their health, finances or relationships because of privacy concerns. Dark aims to address these concerns by combining data privacy with what Eternal describes as high-quality AI responses. The company said it has fine-tuned its models to deliver answers that prioritise truth over user approval, adding that benchmark results would be disclosed later. Goyal also highlighted the potential risks associated with AI companies retaining user conversations, questioning whether such data could eventually be used to influence consumer behaviour or shape users' beliefs. However, not many details about the platform are available yet. “Dark will launch globally, hopefully within 2026,” Goyal said in the letter. With Dark, Eternal has doubled down on its strategy to commercialise AI capabilities developed for its own operations. In February last year, it launched AI-native customer support platform Nugget after building it as an internal tool over three years to handle millions of customer support interactions per month across Zomato, Blinkit and Hyperpure. Nugget offers businesses conversational chatbots and co-pilots, alongside voice AI, image classification and workflow automation to businesses. These disclosures came as Eternal reported a 92% surge in its net profit to ₹125 Cr in the quarter ended September 30, 2026 (Q2 FY27), from ₹65 Cr in the previous year’s quarter. On a sequential basis, profit rose 36% from ₹92 Cr. Operating revenue for the quarter zoomed 78% YoY and 19% QoQ to ₹24,138 Cr. Including other income of ₹351 Cr, total income stood at ₹24,489 Cr. District Faces Higher Risk From Agentic Commerce Apart from outlining its own AI ambitions, Eternal also raised concerns about the impact of agentic commerce on its group companies. Notably, it said its going-out vertical District faces the highest potential disruption from AI agents as they could bypass the platform to discover restaurants, book tables and purchase event tickets, reducing the platform’s role as an intermediary. Goyal classified District as being in the medium-to-high risk zone, warning that AI agents could weaken content-led competitive advantages, including reviews, ratings and curated listings, by synthesising information from across the internet. However, he believes competition could increasingly favour platforms with stronger execution capabilities. Meanwhile, he said that Zomato and Blinkit are relatively insulated from agentic commerce because their businesses rely on complex fulfillment infrastructure rather than discovery alone. According to consumer surveys cited by Goyal, over 65% of Zomato app openings are exploratory, with users browsing an average of 12 menus and more than 100 food images before ordering. He argued that food discovery is an enjoyable activity that conversational AI may not easily replace. While repeat orders and restocking could shift to AI agents, Eternal said reordered carts account for only a low-single-digit share of activity on both Zomato and Blinkit. It also believes in-app AI assistants could cater to these use cases. More importantly, food delivery and quick commerce require real-time coordination between restaurants, delivery partners, inventory systems, pickers and dispatch operations. He argued that agents could initiate transactions, but the platforms managing fulfillment would remain responsible for delivering orders and resolving issues. This comes as consumer companies are increasingly exploring agent-led shopping and payments. In February 2026, Razorpay and NPCI demonstrated agentic payments through Anthropic’s Claude, enabling users in a pilot to order food and groceries from platforms including Zomato, Swiggy and Zepto within AI conversations. Meanwhile, Meesho has developed Vaani, a voice-first, multi-agent commerce system designed to help users discover and compare products conversationally.
document fetched 2026-10-22 03:02 · canonical https://inc42.com/buzz/eternal-unveils-privacy-focused-ai-platform-dark-eyes-global-launch-in-2026