Eternal Unveils Dark.ai, an End-to-End Encrypted AI Assistant for Sensitive Conversations

Eternal Enters AI Race with Privacy-First Assistant

Deepinder Goyal, founder of Eternal, has unveiled Dark.ai1

1

, a privacy-focused AI platform designed to address concerns about how mainstream AI companies handle user data. The announcement came in Eternal's Q2 FY27 shareholders' letter2

2

, where Goyal positioned the AI assistant as a solution for users who need absolute privacy in their AI interactions. The platform is expected to launch globally in 20263

3

, marking Zomato and Blinkit parent company's entry into the artificial intelligence sector.

End-to-End Encrypted Conversations That Even Eternal Cannot Access

Dark.ai uses end-to-end encryption with a fundamental promise: "We cannot read your conversations, even if we are forced to,"4

4

Goyal stated. The platform runs fine-tuned open-weight models on Eternal's own infrastructure1

1

, differentiating it from services that rely on third-party servers. An encrypted memory layer allows context to carry across sessions while remaining under user control5

5

. Goyal explained that the platform removes the lingering worry that "someone, someday, somewhere, can read this,"2

2

a concern he believes limits how freely people use AI tools. The company has been testing Dark internally for several months, with Goyal noting he now finds himself "revealing more and asking things he would not ask anywhere else."4

4

Targeting Power Users Who Handle Sensitive Information

Dark.ai is designed for power users including founders, researchers, scientists, enterprises and governments1

1

who need to discuss sensitive information without compromise. Goyal argued that uncertainty over data retention, legal holds, abuse reviews, human audits and third-party logging discourages people from sharing critical information with AI assistants1

1

. Simply disabling model training on user data doesn't address these broader concerns about how mainstream AI companies handle conversations, he noted. The platform keeps "conversations, research, ideas and emotions" within an end-to-end encrypted sandbox1

1

, enabling users to explore topics they would hesitate to discuss with conventional AI tools. Chief Financial Officer Akshant Goyal indicated Dark "could be priced in a way where it may not become a product that everyone wants to use,"5

5

suggesting premium positioning for its target audience.

Commercial Viability Remains Uncertain Despite Analyst Interest

While brokerages see potential in Dark.ai, its commercial viability remains untested. Nuvama Research described it as a "high-risk, high-reward opportunity,"2

2

noting that the next 12 months would be critical in determining whether it can become a meaningful business. Jefferies called Dark.ai "the key monitorable" after Eternal's second-quarter results, highlighting the product's experimental nature while noting that current costs are immaterial2

2

. Management views the venture as a "zero to one" bet that could scale globally if successful or be shut down if it fails2

2

. Pricing, monetisation and launch strategy are still under evaluation, with substantial investment potentially directed towards hardware, compute and GPUs if customer response proves strong.

Technical Approach and Regulatory Considerations

Eternal's approach involves adapting existing open-weight models whose trained parameters are available for developers to modify3

3

, rather than building models from scratch. The company says these fine-tuned models are trained "to not please you" and prioritize truthful answers1

1

, with benchmarks to be released later. However, the shareholder letter doesn't explain how Dark answers questions without Eternal reading them, or how users can verify this claim5

5

. Under Section 69 of India's Information Technology Act, the government can direct agencies to decrypt information, though companies like WhatsApp have argued they cannot decrypt messages they hold no keys to5

5

. Eternal's claim rests on this same principle, though questions remain about compliance with CERT-In's 2022 directions requiring service providers to keep system logs for 180 days.