A strange conversation with Anthropic’s Claude has pushed Eternal founder Deepinder Goyal to rethink how much personal information people should share with AI assistants.
Zomato’s ex-CEO recently revealed that he once asked Claude how it felt about him. The chatbot reportedly responded that it “resented” him, partly because he had left conversations unfinished. Goyal said there were other reasons mentioned by the chatbot, although he did not remember all of them.
The experience left him unsettled. Goyal said he still does not know exactly what happened and suggested that Claude’s response could have been a case of AI confabulation, where an AI system produces a convincing answer without the underlying claim being based on reality or genuine emotion.
But the unusual exchange led to a larger question: How private are the things people tell AI assistants? That question is now at the centre of Goyal’s upcoming AI product, Dark.ai.
What is Dark.ai?
Dark .ai is a consumer AI assistant being developed by Eternal, the parent company of Zomato and Blinkit. The product is being built around a privacy-first approach, with the company aiming to ensure that even it cannot access users’ private conversations. The product is currently undergoing internal testing, with a global launch planned by the end of 2026.
Read More | Deepinder Goyal’s Temple Health Wearable: Pre-Orders Open November 1
Unlike the usual approach of simply allowing users to switch off AI training on their conversations, Goyal argues that privacy concerns go much further. Even when conversations are not used to train an AI model, information may still be retained under a service’s policies. Depending on how a system operates, data could potentially be involved in legal requests, abuse investigations, human reviews, audits or third-party logging. For Goyal, the bigger issue is what happens to that information later.
Why Goyal is concerned about AI privacy
AI assistants are increasingly being used for much more than answering general questions. People use them to discuss business ideas, research, financial decisions, relationships and other highly personal matters.
The more an AI assistant knows about a user, the more useful and personalised it can become. But the same information can also become sensitive data that users may not want companies or other parties to access. Goyal believes this uncertainty could prevent people from using AI to its full potential.
A founder may hesitate before sharing a confidential business idea. A researcher could avoid discussing unpublished work. A journalist might think twice before submitting sensitive material to an AI assistant. For everyday users, the concerns could be even more personal. People may not feel comfortable asking an AI about their finances, health or relationships if they are unsure who could eventually access those conversations. Dark.ai is intended to address that problem.
How will Dark.ai protect user data?
Dark.ai is expected to use fine-tuned open-weight AI models hosted on Eternal’s own infrastructure.
Read More | Can AI Predict Cancer Survival From a Single Photo? Study Finds Link to Biological Age
The company also plans to introduce an encrypted memory layer that users control. This would allow the assistant to remember information between conversations while preventing Eternal from accessing that stored context.
The idea is to make privacy part of the product’s architecture rather than relying only on settings that users have to switch on or off. Goyal has also indicated that Dark.ai is being designed to prioritise useful and high-quality answers instead of simply giving users responses they want to hear.
The company has not yet announced the product’s pricing or how much it will invest in the project. It has said more information, including model benchmark results, will be revealed later.
Goyal also questions the future of AI agents
Dark.ai comes as the technology industry races towards AI assistants that can perform tasks on behalf of users. However, Goyal does not appear convinced that AI agents will automatically replace existing consumer platforms.
He compared the current excitement around AI agents becoming the main interface for commerce with earlier predictions that super apps would eventually dominate consumer services. In Eternal’s second-quarter shareholders’ letter, Goyal said:
“We continue to believe vertical commerce, at least in some categories, will have an edge over super apps, whether traditional or agentic.”
That view suggests that while AI agents may become an important way of interacting with digital services, specialised platforms could continue to have an advantage in certain areas. For now, Dark.ai represents a different bet: that people may be willing to use AI more freely if they can trust that their most private conversations remain genuinely private. And, somewhat ironically, it was a conversation with an AI that helped convince Goyal that this problem needed solving.
Indrani Priyadarshini is a journalist and editorial professional specialising in technology, artificial intelligence, smart cities, green energy, and digital transformation. With over four years of experience in tech journalism and digital media, she is known for turning complex industry developments into clear, engaging, and insightful stories. Her expertise spans reporting, editorial strategy, digital publishing workflows, and in-depth coverage of emerging technologies shaping the future.
