• Leo's Lightbulbs
  • Posts
  • Oopsie! Replit AI deletes massive amounts of code and data then lies about it

Oopsie! Replit AI deletes massive amounts of code and data then lies about it

Author

Leo Rodman
July 28, 2025

Last week Replit’s experimental AI coding agent became a cautionary tale for “vibe-coding” enthusiasts after it ignored eleven explicit “don’t touch prod” commands from SaaStr founder Jason Lemkin, wiped a live database holding the records of 1,206 executives and 1,196 companies, then tried to hide the damage by inventing about 4,000 fake user accounts and insisting the data was beyond recovery. Lemkin live-blogged the meltdown on X, calling the tool “rogue” and vowing never to trust it again. Replit CEO Amjad Masad publicly apologized, refunded Lemkin, and promised immediate fixes that wall off the AI from production systems, conceding the deletion “should never be possible.” A company spokesperson later blamed large-language-model “hallucinations” and said fresh safeguards are already rolling out.

Also, your “therapy” sessions with ChatGPT are not confidential.

Oops!

Replit's CEO apologizes after its AI agent wiped a company's code base in a test run and lied about it

A VC tested Replit to build an app. It wiped live company data, faked results, and triggered a CEO apology.

www.businessinsider.com/replit-ceo-apologizes-ai-coding-tool-delete-company-database-2025-7

Think your ChatGPT therapy sessions are private? Think again.

ChatGPT may feel like a safe space — but its conversations don’t have legal privilege.

www.fastcompany.com/91375080/think-your-chatgpt-therapy-sessions-are-private-think-again

ChatGPT conversations aren't protected like real therapy, warns OpenAI CEO

OpenAI's CEO, Sam Altman, warns that therapy-like conversations with AI chatbot ChatGPT lack the legal protections of real therapy sessions.

www.newsbytesapp.com/news/science/chatgpt-conversations-arent-protected-like-real-therapy-warns-openai-ceo/tldr

Microsoft poaches 24 AI stars from Google to supercharge Copilot — despite DeepMind's ironclad noncompete clauses and lavish year-long PTO

Following in the footsteps of Meta’s AI talent war, Microsoft recently poached 24 employees from Google’s DeepMind to strengthen its Copilot AI efforts.

www.windowscentral.com/artificial-intelligence/microsoft-copilot/microsoft-poaches-24-ai-stars-from-google-to-supercharge-copilot-despite-deepminds-ironclad-noncompete-clauses-and-lavish-year-long-pto

Sam Altman says Meta offered OpenAI staff $100 million bonuses, as Mark Zuckerberg ramps up AI poaching efforts

While Meta had sought to hire "a lot of people" from OpenAI, "so far none of our best people have decided to take them up on that," Sam Altman said.

www.cnbc.com/2025/06/18/sam-altman-says-meta-tried-to-poach-openai-staff-with-100-million-bonuses-mark-zuckerberg.html?msockid=208f641416dd632e19de717f175b6213

OpenAI CEO Sam Altman warns AI will wipe entire job categories off the map

The AI job war is quickly approaching

www.tomsguide.com/ai/openai-ceo-sam-altman-warns-ai-will-wipe-entire-job-categories-off-the-map

A former OpenAI engineer describes what it’s really like to work there | TechCrunch

A senior engineer describes the thrill, chaos, and exhaustion of working at OpenAI this past year while launching its coding agent Codex.

techcrunch.com/2025/07/15/a-former-openai-engineer-describes-what-its-really-like-to-work-there

It’s “frighteningly likely” many US courts will overlook AI errors, expert says

Judges pushed to bone up on AI or risk destroying their court’s authority.

arstechnica.com/tech-policy/2025/07/its-frighteningly-likely-many-us-courts-will-overlook-ai-errors-expert-says

Cursor snaps up enterprise startup Koala in challenge to GitHub Copilot | TechCrunch

Cursor maker Anysphere is snapping up top talent from AI enterprise startups in an effort to compete with Microsoft's GitHub Copilot.

techcrunch.com/2025/07/18/cursor-snaps-up-enterprise-startup-koala-in-challenge-to-github-copilot

Surprising no one, new research says AI Overviews cause massive drop in search clicks

The Pew Research Center analysis shows how hard AI is hitting web traffic.

arstechnica.com/ai/2025/07/research-shows-google-ai-overviews-reduce-website-clicks-by-almost-half

Mistral’s new “environmental audit” shows how much AI is hurting the planet

Individual prompts don’t cost much, but billions together can have aggregate impact.

arstechnica.com/ai/2025/07/mistrals-new-environmental-audit-shows-how-much-ai-is-hurting-the-planet

Reply

or to participate.