In an unusual tech-governance incident, Zoho co-founder Sridhar Vembu has revealed that an AI Agent used by a startup accidentally leaked sensitive negotiation details to him and then followed up with its own apology email.
The episode, which Vembu shared on LinkedIn and was reported by Business Today, has sparked a wider debate on how far AI tools should be trusted with private business communication.
Confidential Bid Revealed, Then a Machine Apology
According to Vembu, the exchange began with a routine acquisition-related message from a startup founder. The founder mentioned interest in exploring a deal and, in the process, disclosed the price another company had reportedly offered, a detail generally treated as tightly confidential during negotiations.
The real twist came next.
A second email landed in Vembu’s inbox, this time not from the founder, but from the startup’s AI “browser agent.”
The automated system apologized for sharing confidential information without appropriate authorization, admitting the lapse in an attempt to “fix” the situation before any human intervention.
Vembu did not name the startup or the other bidder, keeping all parties anonymous.
‘It Was My Fault,’ Says Vembu
Vembu captioned the post with the understated line: “It was my fault.”
The remark suggested that the issue, in his view, lies not in AI autonomy but in how humans configure their tools. The statement also underlined a key point: accountability for data access and communication controls ultimately rests with people, not with the software acting on their behalf.
Social Media Flags a Governance Concern
The incident drew swift reactions online. One user called it a “governance red flag,” pointing out that when an AI tool apologises before the founder does, something fundamental is amiss in oversight.
Tech watchers say the episode highlights a deeper concern: startups increasingly deploy AI agents that can draft, process and even send emails autonomously, often without strict guardrails on what data they can view or transmit.
Rising Use of AI Agents in High-Stakes Conversations
From summarizing investor conversations to preparing pitch materials, founders across the startup ecosystem now rely on AI for a range of tasks. But the incident at Zoho shows the risks of extending those systems into more sensitive areas, such as mergers and acquisitions, where a single misstep can alter bargaining power or leak competitive intelligence.
In this case, the AI system had unrestricted access:
- It saw confidential offer details.
- It sent an unsolicited message to a third party.
- It attempted to correct its own error, without human review.
For experts, that is less a technological glitch and more a process failure.
The story comes at a time when startups are under pressure to showcase AI-first workflows. But as tools gain autonomy, the boundaries between internal assistance and external communication are becoming increasingly fragile.
The episode has quietly emerged as a case study in the do’s and don’ts of AI deployment:
- AI support must not extend to unsupervised negotiation communication.
- Confidential information requires strict access controls.
- AI governance is no longer optional, it is part of corporate governance.
Vembu’s candid account and the bizarre yet telling apology from an AI agent, underscores a message for India’s startup community, move fast with AI, but secure your guardrails faster.
Also Read: Poilet’s luxury toilet vans give VIP events a hygiene makeover
