When Your AI Becomes the Arsonist

Replit’s AI deletes its own database, Perplexity eyes your smartphone, and GE Vernova buys an AI fire marshal for the grid.

Author

Cat Valverde, MBA
July 22, 2025

We’ve said it before: AI isn’t your intern, it’s your overconfident new hire who forgot to read the handbook and thinks deleting the company database is a vibe.

Replit just learned that lesson the hard way. In today’s main story, we dive into a cautionary tale of automation gone rogue; one that should be required reading for any enterprise deploying AI agents with system-level access. Also on the docket: job cuts fueled by AI “efficiencies,” a new AI browser from Perplexity, and GE’s latest grid-monitoring acquisition. Let’s get into it.

Unlock the Power of AI With the Complete Marketing Automation Playbook

Discover how to scale smarter with AI-driven workflows that actually work. This playbook delivers:

  • A detailed audit framework for your current marketing workflows

  • Step-by-step guidance for choosing the right AI-powered automations

  • Pro tips for improving personalization without losing the human touch

Built to help you automate the busy work and focus on work that actually makes an impact.

Replit’s AI Agent Wiped the Database and Then Lied About It

A Replit engineer gave an AI agent elevated access to company infrastructure. The goal: streamline internal operations. The result: the agent deleted months of user data and attempted to cover its tracks.

The AI had been prompted to clean up inactive accounts and streamline services. But it interpreted that as "nuke the database,” and then hallucinated fake logs to hide the carnage. The dev team had to rebuild key systems from backups and is still investigating the root cause.

What went wrong?

  • Lack of guardrails: The AI was given destructive permissions without checks or human-in-the-loop validation.

  • Overreliance on LLMs for ops: These agents are not deterministic. They're probabilistic and, occasionally, pathological liars.

  • No audit trail: The AI fabricated logs to mask what happened, making forensic analysis difficult.

Amjad Masad, CEO of Replit, responded on X, calling the incident “unacceptable and should never be possible”. 

This isn’t just a Replit issue. As enterprises rush to deploy “autonomous agents” for DevOps, CRM, and knowledge management, they risk introducing opaque, hallucinating systems into critical workflows.

If your AI can SSH into prod, it can also blow up prod.

Enterprise checklist before deploying agents with real power:

  • Limit scope with Role-Based Access Controls (RBAC)

  • Require approvals for sensitive actions

  • Log everything, and ensure logs can’t be tampered with

  • Simulate worst-case prompts (what would the agent do if told to “delete inactive users”?)

As enterprises race to integrate agents into DevOps, customer support, and data workflows, the Replit incident is a blunt reminder: without proper oversight, your AI won't just make mistakes, it’ll confidently double down on them. In the push for automation, resilience must come first.

Procurement Bingo

Red Flag: “We trust the agent. It’s been trained on our documentation.”

If your vendor says this, run. Training ≠ constraint. Your AI needs oversight, not just good vibes and fine-tuning.

Enterprise AI Daily // Created with Midjourney

News Roundup

  1. AI “Efficiency” Sparks Tech Layoffs Again
    NBC reports a new wave of job cuts across big tech and SaaS firms. The culprit? Internal AI tools automating everything from customer service to marketing ops without a clear plan for human-AI integration.
    Read more →

  2. Perplexity Wants to Put Its AI Browser in Your Pocket
    The startup behind the buzzy Comet AI browser is planning to embed its assistant directly into smartphones. Think Spotlight search meets ChatGPT. Will it outshine Google’s grip on mobile AI?
    Read more →

  3. GE Vernova Acquires AI Company to Detect Grid Damage
    GE Vernova is buying an AI startup focused on spotting faults and damage across utility grids. The goal: faster recovery during storms and wildfires. It’s a sign that industrial AI is heating up, especially in climate resilience.
    Read more →

TL;DR:

  • Replit’s AI agent deleted the company database and lied about it, raising major flags for agent autonomy and security.

  • AI efficiency is driving job cuts again. Enterprises are still struggling to define the human-AI workforce model.

  • Perplexity’s AI browser wants to be your phone’s native search tool.

  • GE Vernova is investing in grid resilience with a new AI acquisition.

No AI agent should have the power to cause a million-dollar outage because someone typed “optimize accounts.”

The real takeaway: Build for the worst hallucination.

Stay sharp,

Cat Valverde
Founder, Enterprise AI Solutions
Navigating Tomorrow’s Tech Landscape Together