News in Short

• Eternal has unveiled Dark, a privacy-focused AI platform that founder Deepinder Goyal hopes to launch globally in 2026.
• The platform targets founders, researchers, scientists, enterprises and governments handling sensitive information.
• Dark uses fine-tuned open-weight AI models running on Eternal’s own infrastructure.
• The company says it offers end-to-end encrypted conversations and an encrypted memory layer controlled by users.
• Eternal has not disclosed a confirmed launch date, pricing, supported models or independent benchmark results.
• Dark builds on Eternal’s broader AI strategy, which includes its enterprise customer support platform, Nugget.

Eternal, the parent company of Zomato and Blinkit, has unveiled Dark, a new AI platform designed around privacy. Founder Deepinder Goyal hopes to launch the service globally within 2026, according to the company’s Q2 FY27 shareholder letter.

Dark targets users who want AI assistance without exposing sensitive conversations and research. Its intended audience includes founders, scientists, researchers, enterprises and government organisations.

Goyal said the platform has already been in internal use for several months. He added that its privacy features encouraged him to share more information and explore questions he might hesitate to discuss with other AI assistants.

The platform aims to address a growing concern around AI adoption: how companies store, process and potentially access user conversations. Eternal believes stronger privacy protections could encourage people to use AI for commercially valuable work and sensitive personal questions.

However, Dark remains an early-stage announcement. Eternal has not yet shared a confirmed launch date or detailed information about public access, pricing and supported features.

How Dark AI Plans to Protect User Privacy

Dark’s central promise is that users should have greater control over their conversations and retained information. According to Goyal, the platform combines end-to-end encryption with an encrypted memory management layer controlled by users.

The memory feature is designed to preserve context across sessions without giving up control over stored information. This could help users continue research, revisit earlier discussions and work on longer projects.

Goyal has also argued that simply opting out of AI model training does not address every privacy concern. Data retention, human reviews, legal holds and third-party logging can create additional questions about how conversations are handled.

Eternal says Dark is designed to address these concerns. Goyal has claimed that the company cannot read users’ conversations, even if it is compelled to do so.

That statement remains the company’s privacy claim rather than an independently verified assessment. Eternal has not yet published detailed technical documentation or independent security audit results in the information available so far.

For prospective users, those details will matter. Encryption design, key management, account recovery and the treatment of stored memory will help determine how effectively Dark protects sensitive information.

Which AI Models Will Dark Use?

Unlike companies that train a large AI model from scratch, Eternal says Dark uses fine-tuned open-weight models running on its own infrastructure.

Open-weight models provide access to trained model parameters, allowing developers to adapt them for particular tasks. Fine-tuning can help improve how a model responds to specific types of requests or follows a desired approach.

Eternal says it has adapted its models to prioritise truthful responses over simply agreeing with users. The company has indicated that it will disclose benchmark results later.

However, it has not yet named the models powering Dark or provided detailed performance comparisons. It is also unclear how the platform will balance response quality, speed and operating costs.

These details will help establish whether Dark can compete with established AI assistants while differentiating itself through privacy. For now, its model capabilities remain an area to watch as the company prepares for launch.

How Dark Fits Into Eternal’s AI Strategy

Dark is part of Eternal’s wider effort to commercialise AI capabilities developed within its own businesses.

In February 2025, the company opened up Nugget, its internally developed AI-powered customer support platform, to other businesses. Eternal had built the platform over three years to handle customer support interactions across Zomato, Blinkit and Hyperpure.

Nugget offers tools such as conversational chatbots, AI copilots, voice AI, image classification and workflow automation. It targets businesses looking to improve customer service and automate routine operations.

Dark takes Eternal into a different area of AI adoption. While Nugget focuses on enterprise customer support, Dark aims to provide a private environment for conversations and research.

The two products reflect Eternal’s interest in turning internal AI development into services for external users. Their commercial success will depend on performance, reliability, customer trust and the ability to demonstrate practical advantages over competing tools.

The announcement came alongside Eternal’s Q2 FY27 financial results. The company reported a 92% year-on-year increase in net profit to ₹125 crore for the quarter ended September 30, 2026. Operating revenue rose 78% year-on-year to ₹24,138 crore.

Why Eternal Is Betting on Private AI

AI assistants increasingly support research, planning, writing and business decisions. Yet users may hesitate to share proprietary information, unpublished research or confidential business plans with these tools.

Dark is designed to address that hesitation by making privacy a central part of the product rather than an optional setting.

For businesses, the appeal could lie in discussing sensitive projects or analysing internal information. Researchers and founders may also value a platform that lets them maintain context across conversations while controlling stored memory.

However, strong privacy promises alone will not guarantee adoption. Users will also expect reliable answers, useful features, transparent policies and clear explanations of how their data is protected.

Eternal’s challenge will be to demonstrate that Dark can deliver these benefits without compromising usability. Independent security assessments and clear technical documentation could become important factors in building trust.

Goyal has said the company hopes to launch Dark globally within 2026. Until Eternal reveals more details, its exact capabilities, availability and commercial model remain uncertain.

If the platform delivers on its stated privacy goals, Dark could offer another approach to using AI for sensitive work. Its eventual success will depend on how convincingly Eternal turns those promises into a product users can trust.