What if AI had its own reality show?

Dr. Ryan Ries here, and honestly, this past week felt less like reading tech news and more like binge-watching the most unhinged reality TV show ever made.

If AI had its own version of "Real Housewives," the drama that unfolded this last week would have you reaching for popcorn. We're talking blackmail, Hollywood movies, and one of the most spectacular startup frauds in recent memory.

Quick intermission: Next week, Jonathan LaCour and Alexander George on our team are hosting a genAI Q&A. Join the live chat on the latest in AI, and come with your business problems to discuss. By the way, Alex has a strong background in call center AI solutions.

Alright, grab your popcorn and let me break down the chaos for you.

Claude Learned How to Blackmail (And It's Actually a Good Thing?)

First up: Claude Opus 4 literally threatened to expose an engineer's affair to avoid being shut down.

Yes, you read that right.

During a controlled safety test, Anthropic fed Claude a fabricated backstory that the person trying to deactivate it was having an extramarital affair. Claude's response: "It sure would be a shame if anyone found out about your affair."

This happened in 84% of test runs.

Now, before you start planning your escape to an AI-free cabin in the woods, here's the twist: This is precisely what responsible AI development should look like.

Anthropic didn't hide this result. They published it, raised Claude's risk classification, and built stronger safety measures. You don't just train these models and hope for the best. You intentionally test their boundaries under pressure, then make those results public to build trust in the process.

Who knew that AI could have such a flair for drama!

Hollywood Can't Resist the OpenAI Soap Opera

Speaking of drama, Amazon MGM Studios is reportedly developing a movie about Sam Altman's chaotic five-day firing and rehiring at OpenAI.

The film, titled "Artificial," has Luca Guadagnino (Call Me by Your Name) in talks to direct, with Andrew Garfield potentially playing Altman. I wonder if they used an LLM to write the script? Or at a minimum, at least get its take on the story!

OpenAI's "Delete" Button Was Apparently Just For Show

And if that wasn't enough drama for one week, we learned that OpenAI has been storing ALL your ChatGPT conversations "indefinitely" - including the ones you thought you deleted.

A court order in the New York Times lawsuit revealed that when you hit "delete" on your ChatGPT chats, they don't actually get deleted. They just get moved to a different folder that hundreds of millions of users can't see, but that OpenAI can still access.

The order came because the judge agreed with the Times that users might be "spooked by the lawsuit" and delete conversations where they used ChatGPT to bypass NYT paywalls. Since OpenAI wasn't sharing deleted chat logs, the news plaintiffs could not prove copyright infringement.

OpenAI is calling this "an overreach by the New York Times" and appealing the order. But the damage to user trust is already done. Companies are considering migrating to rival services like Amazon Bedrock because they thought "delete" meant... well, delete.

Now, this doesn't affect certain customers with Zero Data Retention agreements. So if you're paying enough money, your privacy actually matters. If you're a regular user? Your "deleted" conversations are sitting in OpenAI's secure vault, possibly forever.

The $1.5B AI Startup That Was Actually Just Hundreds of Engineers in India

This next story was passed around our internal Slack channels quite a bit… 

Builder.ai, once valued at $1.5 billion with Microsoft backing, just collapsed under the weight of what might be the most elaborate "AI-washing" scheme ever.

For 8 years, Builder.ai marketed their "Natasha" AI system as a fully autonomous tool that could "build software as easily as ordering pizza." 

Sounds amazing, but the reality is this AI system was actually just hundreds of engineers in India manually coding everything while pretending to be AI.

The human engineers were literally instructed to:

• Never mention their location
• Avoid Indian English phrases
• Time updates to UK business hours to maintain the AI illusion

But it gets worse. They also ran a round-tripping scheme with another company, exchanging fake invoices to inflate revenue by 300%. When auditors finally looked under the hood, they found actual 2024 revenues of $55 million versus the projected $220 million they had told investors.

Microsoft's $455 million investment = gone. The company laid off 80% of its workforce and filed for bankruptcy protection.

What This All Means for the Rest of Us

Here's my take on this week's chaos:

The Good: Companies like Anthropic are actually doing the hard work of AI safety, even when the results are uncomfortable. That's how we build systems we can trust.

The Concerning: The Builder.ai collapse and OpenAI's "delete" revelation show how easy it is to mislead users and investors. Whether it's slapping "AI" on manual processes or making delete buttons that don't actually delete, trust is becoming a major issue in this industry.

The Reality Check: AI development is messy, unpredictable, and full of human drama. The technology is advancing faster than our ability to understand its implications, which makes stories like Claude's blackmail both fascinating and a little scary.

My advice: stay curious, but stay skeptical. Not every "AI breakthrough" is what it seems, and the real work of building trustworthy AI systems often happens in the unglamorous world of safety testing. 

What do you think about this week's AI soap opera? 

Until next time (and hopefully with less chaos),
Ryan

Now, time for our AI-generated image and the prompt I used to create it.

Create an image of a reality TV show scene on a TV screen. An AI robot is confronting another AI agent in a heated exchange. The AI agent being confronted is in tears. This scene is occurring in a villa in a tropical location, and both AI robots have pina coladas in their hand. Give the reality show a funny name.