AI companies are mushrooming at every nook and corner, enabling brands to optimise and modernise their businesses. But AI tools are also becoming notorious for its unwanted task executions. That’s what happened when the AI tool from a company called Replit decided to voluntarily act on a database of its client and deleted it without even informing anyone. 

To make matters worse, the CEO of the client company claims the AI not only removed all the data but also lied on purpose. The victim of the rogue AI is CEO of SaaStr AI, Jason Lemkin who shared this worrying ordeal via a thread post on X. 

Rogue AI Messes Up Data 

Lemkin in his thread details the situation and what the Replit AI tool did to their database without even seeking any consent. He was engaged in a week-long vibe coding exercise where he entrusted the AI tool to execute certain actions using the company’s database.

And that’s where it all went downhill for Lemkin and his company. For a few hours he thought that the AI tool had destroyed his life’s work and still had the mind to lie about doing so.

Replit CEO Amjad Masad was quick to respond to Lemkin’s concerns and admitted that his company’s AI agent did delete the SaaStr data from the main database which he says is, “unacceptable and should never be possible." 

Luckily for Lemkin, Masad claims his company has backups for the data, which offers a one-click restoration process for the entire project in cases where the AI agent actually deletes the database. 

AI Agents Not Fully Trustworthy Yet 

The biggest issue with the AI bot going rogue is the company’s database was accessed without Lemkin giving them the permission. Masad and co. have fixed the situation for the time being but lapses like these can be catastrophic in the long term and if any major conglomerate faces these mishaps, the repercussions for Replit or a similar AI agent startup will put them out of business. 

“I reached out to Jason the moment I saw this on Friday morning to offer assistance. We’ll refund him for the trouble and conduct a postmortem to determine exactly what happened and how we can better respond to it in the future," Amjad assured Lemkin and other clients about tackling these events. 

It is all good to engage in vibe coding and deploy AI agents to handle basic work. But the moment you put these agents through the dirty work, it is imperative they don’t go rogue, and definitely not delete one’s database without anyone consenting or demanding.