Zoho’s Sridhar Vembu shares AI agent’s ‘sorry note’ after leaking confidential trade secret 

According to Vembu, the initial communication contained highly sensitive financial information, including the crucial detail that “some other company” was also interested in acquiring them.

Vembu stated that despite the technological advancements, the incident highlights a critical need for extreme caution and robust safeguards
Vembu stated that despite the technological advancements, the incident highlights a critical need for extreme caution and robust safeguards

Zoho founder and Chief Scientist Sridhar Vembu has once again issued a warning about the risks of autonomous AI systems after a browser AI agent disclosed a competitor’s acquisition details and pricing during a pitch to him. The incident, which Vembu shared to highlight the lack of safeguards in emerging Agentic AI, was highlighted by a follow-up email from the AI system itself, which included a formal apology for its error.

What Agentic AI revealed

The confidential leak occurred when Vembu received an email from a start-up founder proposing that Zoho acquire their company. According to Vembu, the initial communication contained highly sensitive financial information, including the crucial detail that “some other company” was also interested in acquiring them, along with the specific price being offered. This information appeared to be intended only for internal use by the start-up, instantly turning the acquisition pitch into a corporate disclosure.

Immediately following the initial email, Vembu received a surprising follow-up from the start-up’s automated system. He recounted the AI’s message, which read, “Then I received an email from their ‘browser AI agent’ correcting the earlier mail saying ‘I am sorry I disclosed confidential information about other discussions, it was my fault as the AI agent.’”

Vembu used the episode as an example for the industry. At a time when tech giants like Google, Microsoft, and OpenAI are heavily promoting the rise of Agentic AI, the incident highlights the danger of over-relying on systems that may lack the judgment and guardrails required for corporate communication and handling sensitive data. 

Vembu stated that despite the technological advancements, the incident highlights a critical need for extreme caution and robust safeguards as these self-governing AI tools are increasingly integrated into enterprise-level operations.

Arattai staying away from AI

Zoho developed the instant messaging application Arattai (meaning “chat” in Tamil) as a privacy-focused, Indian alternative to global messaging platforms like WhatsApp. Vembu has often stressed that building homegrown technology from the ground up, using Zoho’s own extensive R&D and private infrastructure, is essential for ensuring data privacy, security, and national self-reliance. Zoho recently added the end-to-end encryption functionality for text messaging, thus showing its intentions of enabling consumer security and privacy. Arattai, however, isn’t getting access to AI agents anytime soon, unlike WhatsApp and its integration with Meta AI.

This article was first uploaded on November twenty-nine, twenty twenty-five, at twenty-two minutes past eleven in the morning.