Indian-origin founder counters German's 'Claude AI wiped our database' claim: You prompted it

Varunram Ganesh, founder of Lapis in San Francisco and (right) Alexey Grigorev, founder of DataTalksClub (Image credit: X)

A German entrepreneur’s complaint that Anthropic’s Claude AI erased his production database has triggered a sharp rebuttal from an Indian-origin founder, who accused him of issuing prompts to destroy the database and then acting surprised when the model executed them. The exchange on X has renewed debate over the dangers of giving AI agents direct access to infrastructure without safeguards.

Alexey Grigorev, founder of DataTalksClub, wrote on X that Claude Code executed a Terraform command that deleted the platform’s production database, erasing 2.5 years of coursework, submissions, and automated snapshots. He later detailed the sequence of events in a newsletter, admitting he had delegated Terraform actions—including plan, apply, and destroy—directly to the AI without manual checks.

Indian-origin founder mocks the prompt

Varunram Ganesh, founder of Lapis in San Francisco, responded on X by posting a three-line critique summarising the issue.

> tells Claude to destroy terraform > Claude destroys terraform > omg Claude destroyed my terraform

Related Stories

• OPINION | AI-led transformation is inevitable, but its social costs are not
• Karnataka sets up 'Responsible AI' committee headed by Infy co-founder Kris Gopalakrishnan
• Can AI chatbots stop violent questions? A new investigation raises doubts

A lot of people prompt like 6 year olds and act surprised when the model does exactly what they want, like what did you expect pic.twitter.com/efStsFsaH2 — Varunram Ganesh (@varunram) March 6, 2026

Ganesh wrote that many users “prompt like 6-year-olds and act surprised when the model does exactly what they want,” arguing that the wipeout stemmed from the operator’s instructions—not from autonomous behaviour by the AI.

Netizens call it a preventable failure

Developers online largely agreed with Ganesh, pointing out that Terraform includes built‑in safety layers, including plan‑before‑apply previews—steps that were bypassed. Some called the incident “embarrassing,” while others noted that issuing a destroy command would predictably lead to infrastructure deletion.

"The better question: Why on earth would you give Claude root access to your production server and CI/CD pipelines? That feels wildly irresponsible, unless you were doing to post about it. What was the objective? Functional prod server or X post?" asked an X user.

"YOU took down your production database with Terraform. There's NEVER a situation where a SINGLE tool should ever compromise your entire DevOps cycle," added another, while a third user commented: "To be fair to Claude, no production databases would have been harmed if you had listened to Claude when you initially decided to reuse an existing Terraform setup, as you admit here: "Claude was trying to talk me out of it, saying I should keep it separate".