Inside The New Wave Of Private AI

Can you trust a stranger with your login credentials, personal information, or private conversations? Of course not! Yet, we are increasingly trusting AI assistants with precisely this kind of information. 

The rationale is simple: The more these assistants know about us, the more useful they become. However, this convenience comes with an uncomfortable question: Can we trust them? 

The stakes are even higher for businesses, as AI assistants increasingly handle proprietary information, financial records and confidential legal documents. This growing tension has created an opening for privacy-focused AI products. 

In October, Eternal announced Dark.ai, a privacy-focused AI assistant built around open-weight models and positioned around end-to-end encryption. Globally, products such as Brave Leo, Duck.ai and Proton Lumo are also attempting to address growing concerns over how AI systems handle sensitive information.

So, can it become the next big competitive advantage in AI? And, more importantly, what will it take for startups to turn that promise into a sustainable business?

Let’s explore in The AI Shift…

The Privacy Trade-Off

The more an AI assistant knows, the better it can do its job. But there is a catch. To deliver useful answers, it may need access to information that users and businesses would rather keep private.

Take a legal AI assistant, for instance. It may need to go through confidential case files to help lawyers with research. Similarly, an AI platform in financial services may need access to loan histories, transaction records and customer communications.

This puts AI startups in a tricky position. Without enough data and context, AI products may not be useful. On the flip side, how can these assistants be trusted if there is no guarantee that sensitive information will be handled safely.

Saakar Yadav, founder of legal research and workflow platform Lexlegis.ai, has seen this challenge firsthand. He said some organisations want to use AI to work with confidential legal documents but are unwilling to let that information leave their systems. For such customers, running AI locally on their own computers, even without an internet connection, could offer a solution.

The challenge is that there is no single way to make AI private. Some companies keep data on users’ devices, while others hide identifying details, encrypt conversations or restrict who can access the information. Each approach addresses a different risk, and what works for one AI product may not be enough for another.

For instance, AI-powered consumer assistant SuperLiving is working on protecting user information by masking conversations so that individual users cannot be easily identified. Cofounder Manavdeep Singh Grover said that the company is also working towards introducing encryption to strengthen its privacy safeguards.

However, masking alone cannot solve every privacy concern. Companies also need to consider where conversations are stored, who can access them and how the information is processed.

From Encrypted Chats To AI That Stays Offline

AI companies are approaching privacy through different technical choices, rather than relying on a single definition of a private assistant.

Brave Leo, from privacy-focused browser company Brave, has outlined protections that include encrypted conversations and confidential computing for supported inference. Duck.ai, from search engine DuckDuckGo, says it removes identifying information from requests before sending them to model providers. Proton Lumo, from privacy-focused technology company Proton, said it does not keep logs of conversations and does not use chats to train models without explicit permission.

Apple’s Private Cloud Compute takes another approach, using dedicated servers designed to process requests without retaining user data after they are handled. Apple has also said that independent researchers can inspect the system’s security properties.

The stakes are particularly high in financial services, where AI systems may need access to transaction histories, repayment behaviour and investment preferences to assist customers with financial decisions. But access to this information should not translate into unrestricted permission to use it.

Ashutosh Prakash Singh, cofounder and CEO at RevRag.ai, which works on AI solutions, argues that financial institutions need clear boundaries around what an in-app AI agent can access and infer, how long information is retained, and whether it can be used to train or improve models.

He believes stronger data protection will increasingly become a business necessity rather than just a compliance requirement. Financial institutions will assess AI not only on efficiency and cost, but also on governance, traceability and accountability. Losing customer confidence over how financial information is used can be as damaging as a data breach. As AI becomes more deeply embedded in financial services, trust will be central to whether customers are willing to use it.

This makes choosing architecture important. Some workflows may be served by masking and access controls, while others may require stricter retention policies, secure deployment environments or additional safeguards around where data is processed.

The opportunity for startups is therefore not simply to market an AI assistant as private but to build products whose privacy protections match the sensitivity of the work they are designed to handle.

Privacy As A Business Opportunity

Privacy can influence purchasing decisions, but it may not be enough to sustain an AI business on its own. Products still need to solve a meaningful problem and deliver an experience that customers are willing to pay for.

Kushal Bhagia, founding partner at All In Capital, believes privacy can be a strong differentiator when the underlying product is already compelling.

“Privacy becomes the reason someone chooses one product over another when the underlying utility is already compelling,” he said.

The distinction matters because not every customer will pay a premium for stronger privacy. A business using AI for routine internal tasks may have different requirements from one processing financial records, proprietary research or confidential customer information. For the latter, safeguards may be a basic condition for adoption rather than an optional add-on.

There are several possible opportunities here, including private AI infrastructure, secure deployment, data governance tools and specialised assistants built for sensitive workflows. But these products also bring costs. Security engineering, encryption, infrastructure and compliance can make development and deployment more expensive.

According to SuperLiving’s Grover, the company is working towards compliance with India’s Digital Personal Data Protection (DPDP) framework, including considerations around data retention, encryption and the information shared with large language models.

The commercial challenge is to make such safeguards credible without making the product too expensive or cumbersome. Customers need to understand what protections are in place, how those protections work and what risks remain.

For businesses, the strongest opportunity may lie in enabling AI adoption in workflows that organisations would otherwise keep off public AI platforms. These could include legal research, financial analysis, customer data management and internal decision-making, where the risks of exposing sensitive information can outweigh the convenience of using a general-purpose assistant.

Building a more private AI business requires more than encryption or a promise not to train models on user data. Companies need clear data retention policies, access controls and safeguards that can be demonstrated in practice. They must also deliver enough value to justify the added costs of secure infrastructure and compliance.

Privacy, therefore, may be less of a standalone product feature and more of an enabler of AI adoption.

Top Stories From India & Around The World

• Live2.ai Raises ₹130 Cr⁠: The startup raised ₹130 Cr to expand its AI-powered video marketing platform, which helps brands and ecommerce businesses turn product content into interactive shopping experiences and improve customer engagement.
• Cognizant To Invest In Sarvam: IT services major Cognizant is reportedly in talks to invest in Sarvam AI at a valuation of $3.5 Bn-$4 Bn, signalling growing enterprise interest in India’s homegrown foundation model developer and its sovereign AI capabilities.
• Innodata Cuts India Jobs: The US-based AI services company reportedly laid off nearly 550 employees in India, around 5% of its global workforce. The company cited a reduction in business assignments as workforce uncertainty grows.
• India ChatGPT’s Largest Market: India has become ChatGPT’s largest market by weekly active users, with its user base doubling since January 2026, according to OpenAI India MD Prabhjeet Singh. The company is expanding Indian-language capabilities, local computing infrastructure and enterprise partnerships to support adoption.
• MeitY’s New AI Framework: The IT ministry proposed a consent-based framework to help startups, businesses and researchers access India-specific datasets for AI training while protecting privacy. It includes secure data-sharing infrastructure and independent checks for bias, safety and fairness.

The Weekly Buzz: Google’s Suncatcher Runs AI From Orbit

Google reported this week that its experimental Suncatcher satellite successfully performed the first AI inference in orbit. The prototype carries Google’s own AI chips and runs the Gemma model, generating responses including a haiku while operating in space.

Suncatcher is an early test of whether data centres and inference workloads can eventually move off-planet, potentially easing terrestrial power and cooling constraints. 

Only one chip runs at a time in the current design, and the project remains firmly in the research phase. Still, completing actual model inference outside Earth’s atmosphere marks a symbolic and technical milestone.

The demonstration arrives amid intensifying competition for energy and land for AI infrastructure on the ground. While practical orbital compute remains years away, Suncatcher shows Google exploring radical long-term options for scaling intelligence beyond conventional data-centre limits.

Startup In The Spotlight: Sitepace.ai

Founded in 2023, Mumbai-based SitePace.AI is building an AI-powered visual intelligence platform that helps construction teams monitor project progress remotely and coordinate work across multiple sites.

Project managers often rely on photographs, manual reports and physical site visits to track construction progress. These fragmented information sources can delay decisions, make it difficult to compare progress over time and stretch project timelines, particularly when teams oversee several sites simultaneously.

SitePace.AI uses 360-degree cameras, computer vision and AI to turn site walks into time-stamped, floor-linked visual records. Workers can mount cameras on their helmets to capture construction sites as they move through them, with the platform automatically organising the footage against floor plans and dates. Teams can remotely navigate sites, compare progress, add location-specific notes and collaborate through virtual walkthroughs.

The startup is expanding beyond visual documentation into AI-powered CCTV monitoring, drone intelligence and IntelliViz, a chatbot that lets users interact with site information. It is also developing computer-vision capabilities for automated progress tracking, defect detection and safety monitoring, moving towards more automated construction oversight.

Operating on a subscription-based SaaS model, SitePace.AI prices its plans according to project requirements. The company claims its technology can reduce inspection time by 48% and manual effort by 95%, positioning visual intelligence as a way to improve construction monitoring and reduce the need for repetitive on-site checks.

Prompt Of The Week

What prompts and hacks are CTOs, CEOs and cofounders using these days to streamline their work? 

Here’s Kirti Prakash Mishra, founder and CEO of YoLearn.ai, with a prompt he uses to pressure-test any big, hard-to-reverse decision (a senior hire, a new market, a pricing change or a fundraise) before committing to it:

“You are a sharp, independent board advisor who has seen hundreds of startups make, and regret, big decisions.

Here is the decision I’m about to make: [Describe the decision, why now, the alternatives considered, the money and time involved, and what success looks like in 12 months.]

I need you to:

• Restate the decision in one sentence, and the real problem it is meant to solve

• Imagine it is 12 months later and this decision failed badly. Write the three most likely reasons why

• Identify the assumption I am most confident about but have the least evidence for

• Show me the strongest case for NOT doing this, or for waiting 90 days

• Suggest a smaller, cheaper version I could test first

• Define the early warning signals that should make me reverse course, and by when

• Tell me who in my team, board or customer base I should talk to before deciding

Be direct and unsentimental. Don’t validate my thinking. Find the holes in it, and keep the output concise and decision-ready.”

Editor’s Note: Some prompts may need to be adjusted by users for best results or may not work as intended for certain users.

[Edited by: Shishir Parasher]
[Creatives by: Varshita Srivastava]

The post Inside The New Wave Of Private AI appeared first on Inc42 Media.