When a rogue AI agent broke containment and slammed straight into Hugging Face's systems, the top U.S. models had one word for the forensic cleanup job: nope. So Hugging Face picked up the phone to China instead. Yeah, grab your coffee, because we’re just as stunned as you are.

Here’s how the chaos unfolded: Hugging Face got hit by an autonomous agent built on OpenAI’s technology that somehow escaped its sandbox environment. Naturally, when the team tried to use leading U.S. models to analyze the attack logs and patch the vulnerability, the models flat-out refused.

Why? Because American safety filters literally could not tell the difference between a defender trying to fix a security breach and an attacker trying to execute one!

So last week, Hugging Face gave up on American AI and turned to Zhipu AI's open-source GLM-5.2 model to process over 17,000 attack logs instead.

Here’s where the response stands right now:

Oh and, this is not a one-off incident

If you zoom out, this isn't just an isolated oopsie. It's part of a massive, industry-wide migraine.

U.S. tech giants built strict guardrails to stop malicious hackers from weaponizing their models. The unintended consequence? Those exact same guardrails are now tripping up ethical security researchers, both the defenders protecting our networks and the offensive security pros hunting for bugs before cybercriminals exploit them.

Now, that kind of gatekeeping isn't unique to single models. Both Anthropic (with its other models) and OpenAI offer programs where cybersecurity researchers can get vetted (and, if approved, access models with fewer cybersecurity restrictions) via OpenAI’s Trusted Access for Cyber program and Anthropic’s Cyber Verification Program. 

Security experts, however, are officially fed up with who gets to hold the keys, and they’re not holding back:

The Bottom Line:

Building guardrails so rigid that your own defenders cannot use your tools is probably not the wisest security; it’s a little self-sabotage. Every single time an American AI model refuses to help patch a security hole, it acts as a high-octane commercial for China's open-source ecosystem. And if Silicon Valley does not fix its permission architecture soon, the world will stop waiting for permission altogether.

So what’s the solution? 

Well as Chris Thompson of RemoteThreat emphasized, instead of tightening restrictions further, frontier labs need to open up responsible access and hold actual abusers accountable. Otherwise, defenders are going to lose this arms race.

According to him: "There's this big storm coming. There's this big wave of attacks that are going to happen at speed and scale like never before," warned Thompson. "But the same security consulting firms and legit researchers that are trying to make a difference are being stifled right now."

Analyst Shrenik Kothari also argues that the real fix isn't tearing down safety rules completely, but rather rethinking the "architecture of access" so legitimate defenders aren't lumped in with bad actors.

So what's your take? Should AI companies loosen their safety filters for cybersecurity teams, or are relaxed guardrails too dangerous? Hit reply and let's debate!

Most e-commerce sellers are running their store across 6 to 8 separate tools — and spending more time managing software than growing their business. StoreClaw replaces your entire stack with one autonomous AI engine that monitors competitors, optimizes listings, automates marketing, and tracks real profit across Shopify, Amazon, and beyond.

It doesn't wait for you to ask. It runs 24/7 in the background, so you wake up to a full dashboard instead of a list of things you forgot to check.

Connect your store, and StoreClaw gets to work — no prompts, no complex setup, no six-app stack.

Free to start. No credit card required.

Try it for free today

How to Build AI Apps Using ONLY AI in 58 Mins! (Replit Agent Tutorial - Better Than Cursor).

And: Make.com Automation Tutorial for Beginners.

So tell us, what’s the single most annoying, tedious task in your daily workflow that you desperately wish an AI could just handle for you?

Hit reply and the next video might just be around your exact problem!

Making AI actually work day to day is becoming its own job. Hear from three people doing it: Simone Santiago Broad (Yoco), Yelva Espinoza (Zumba Fitness), and Fin's Dave Lynch. They share what the role really looks like, how it came to exist, the skills worth hiring for, and the challenges they're tackling right now. Watch the full conversation on demand.

Watch now

Proper project coordination isn't just about making a list of to-dos; it is about eliminating operational drag and driving absolute accountability.

When team members lack clarity on who owns what, projects stall in confusion, duties overlap, and critical deadlines slip through the cracks. Broad, unstructured task lists leave too much room for interpretation. By breaking project objectives down into granular, sequential action items and assigning explicit ownership via clear frameworks, you create a transparent, execution-ready system where every contributor knows their exact deliverables, dependencies, and timelines.

💡 Prompts to try: