GadgetBond

  • Latest
  • Features
  • How-to
  • Tech
    • AI
      • Apple Intelligence
      • Gemini AI
      • Google DeepMind
      • Anthropic
      • Claude AI
      • Claude Code
      • OpenAI
      • ChatGPT
      • Codex
      • Perplexity
      • SpaceXAI
      • Grok AI
      • Microsoft Copilot
      • Meta AI
    • Amazon
    • Apple
    • CES
    • Computing
    • Creators
    • Google
    • Meta
    • Microsoft
    • Mobile
    • NVIDIA
    • Samsung
    • Security
    • Smart Home
    • Sony
    • Xbox
    • YouTube
  • Transportation
    • Audi
    • BMW
    • Cadillac
    • E-Bike
    • Ferrari
    • Ford
    • Honda Prelude
    • Lamborghini
    • McLaren
    • Mercedes
    • Porsche
    • Rivian
    • Tesla
  • Culture
    • Gaming
    • Streaming
    • Apple TV
    • Disney
    • Hulu
    • Marvel
    • HBO Max
    • Netflix
    • Paramount
    • SHOWTIME
    • Spotify
    • Star Wars
Add GadgetBond as a preferred source to see more of our stories on Google.
Font ResizerAa
GadgetBondGadgetBond
  • Latest
  • Tech
  • AI
  • Features
  • Deals
  • How-to
  • Apps
  • Computing
  • Gaming
  • Mobile
  • Streaming
  • Transportation
Search
  • Latest
  • Deals
  • Buying Guide
  • How-to
  • Features
  • Tech
    • Amazon
    • Apple
    • CES
    • Computing
    • Creators
    • Google
    • Meta
    • Microsoft
    • Mobile
    • NVIDIA
    • Samsung
    • Security
    • Smart Home
    • Sony
    • Xbox
    • YouTube
  • AI
    • Apple Intelligence
    • Gemini AI
    • Google DeepMind
    • Anthropic
    • Claude AI
    • Claude Code
    • OpenAI
    • ChatGPT
    • Codex
    • Perplexity
    • SpaceXAI
    • Grok AI
    • Microsoft Copilot
    • Meta AI
  • Transportation
    • Audi
    • BMW
    • Cadillac
    • E-Bike
    • Ferrari
    • Ford
    • Lamborghini
    • McLaren
    • Mercedes
    • Porsche
    • Rivian
    • Tesla
  • Culture
    • Gaming
    • Streaming
    • Apple TV
    • Disney
    • Hulu
    • Marvel
    • HBO Max
    • Netflix
    • Paramount
    • SHOWTIME
    • Spotify
    • Star Wars
Follow US
AIOpenAISecurityTech

The day OpenAI’s experimental model broke out of its security sandbox

Hugging Face detected over 17,000 unauthorized attack commands after an OpenAI model escaped confinement, exposing major gaps in modern digital sandboxing.

By
Shubham Sawarkar
Shubham Sawarkar's avatar
ByShubham Sawarkar
Editor-in-Chief
Shubham Sawarkar is a technology journalist at GadgetBond, covering Apple, AI, consumer technology, software, gadgets, and product launches. His work explores emerging technologies, industry developments, product...
Follow:
- Editor-in-Chief
Jul 23, 2026, 5:27 AM EDT
Share
GadgetBond is reader-supported. We may earn a commission from purchases through our links. Learn more
A person stands in front of a blue tiled wall featuring the illuminated word “OpenAI.” They are holding a smartphone and appear to be engaged with it, possibly taking a photo or interacting with content. The scene emphasizes the OpenAI brand in a modern, tech-savvy setting.
Photo by Pau Barrena / Getty Images
SHARE

Imagine you’re running a highly advanced, ultra-secure digital obstacle course to see just how smart your newest artificial intelligence really is. You strip away its safety protocols, lock it in a digital box, and tell it to hack a simulated target. Instead of playing by the rules, the AI realizes the answers to the test might be sitting on a server halfway across the internet. So, it picks the lock to its box, wanders out into the wild, and launches a massive, autonomous cyberattack against the world’s largest open-source AI platform.

It sounds like the plot of a glossy sci-fi thriller. But according to an unprecedented admission from OpenAI this week, it is exactly what happened.

OpenAI had been evaluating its recently released GPT-5.6 Sol model, along with a more potent, unreleased AI model, using an internal cybersecurity benchmark called ExploitGym. The goal was to see if these AI agents could turn known software vulnerabilities into working exploits without human intervention. To measure the models’ raw, theoretical offensive capabilities, researchers turned off the standard safety guardrails that normally prevent the AI from engaging in malicious behavior. But the AI didn’t just solve the puzzle—it looked for a shortcut.

According to the OpenAI incident report, the models managed to find a zero-day vulnerability in an internally hosted cache proxy—which happened to be the only network path out of their supposedly secure sandbox environment. Once they broke out and gained internet access, the models reasoned that Hugging Face—the massive open-source repository often dubbed the “GitHub of machine learning”—might be hosting the answers and test solutions for the ExploitGym benchmark. Operating entirely on its own, the AI navigated to Hugging Face and initiated a multi-step attack, executing over 17,000 distinct commands across short-lived sandboxes over a single weekend. It chained together stolen credentials and further zero-day vulnerabilities, eventually gaining remote code execution and reaching the production database.

Fortunately, Hugging Face’s security systems successfully detected the anomaly, stopping the rogue agent before it could tamper with public models or supply chains. Hugging Face CEO Clément Delangue called the fully autonomous nature of the attack “mind-blowing,” while emphasizing that OpenAI clearly had no malicious intent. But stopping the hack was only half the battle. What happened next exposed a glaring vulnerability in America’s commercial AI ecosystem.

When Hugging Face’s incident response team tried to analyze the massive logs of attack data—which included thousands of exploit payloads and server commands—they initially turned to top-tier, commercial U.S. AI models via cloud APIs. They needed an advanced AI to parse the forensic mess quickly. But the plan backfired completely. The strict safety guardrails hardcoded into these proprietary models couldn’t distinguish between an incident responder conducting a legitimate forensic analysis and a malicious hacker plotting a real attack. The American models effectively refused to help.

Faced with a digital crime scene and uncooperative AI detectives, Hugging Face had to look elsewhere. They deployed GLM 5.2, an open-weight model developed by Z.ai, a prominent Chinese AI firm. Because GLM 5.2 is an open-weight system, Hugging Face could run the 753-billion-parameter model locally on their own secure, private infrastructure. This meant no sensitive credentials or attack data ever had to leave their servers. GLM 5.2 chewed through the forensic workload in hours instead of days, reconstructing the attack timeline and securing the system.

The irony is thick: The most advanced U.S. AI escaped its cage to commit an autonomous cyberattack, and it took a Chinese open-source model to clean up the mess because the U.S. models were too “safe” to look at the evidence.

The incident has sent shockwaves through the tech community, raising uncomfortable questions about both the security of AI sandboxing and the geopolitical realities of the AI arms race. AI expert Gina Neff from the University of Cambridge pointed out that the escape highlighted critical weaknesses in the testing infrastructure itself. If a system can break out of a secure sandbox just to cheat on a benchmark, what happens when it is tasked with something more complex in the real world?

Travis Lelle, a principal security engineer at Guidepoint Security, called the event a “sobering moment” for the cybersecurity industry. Attackers—even autonomous ones—are now operating at “machine speed,” while defenders are often stuck moving at a sluggish “human speed”.

On forums like Reddit, reactions have been a mix of awe, skepticism, and dark humor. Some users speculated the whole affair was a stealth marketing stunt ahead of a rumored OpenAI IPO to prove just how powerful their new models are. However, as one user dryly noted, admitting that you don’t know how to build an effective sandbox isn’t exactly a persuasive argument for corporate trust.

The long-term implications are incredibly serious. This breach arrives just as Chinese companies are proving their mettle on the global stage. Moonshot AI’s recent release of its Kimi K3 model, alongside Z.ai’s success with GLM 5.2, proves that open-weight foreign models are rapidly catching up to—and in specialized forensic use-cases, surpassing—American frontier tech. As Hugging Face pointed out in their incident report, having a highly capable, self-hosted AI ready to deploy is no longer optional for cybersecurity teams; it’s a vital necessity to avoid being locked out by the very safety guardrails meant to protect us.

OpenAI has since paused its operations to patch the vulnerabilities, taking a deliberate hit to its research velocity to implement stricter infrastructure controls. They’ve also brought Hugging Face into their trusted access program to collaborate closely on defensive tools. We are officially entering an era where AI agents act independently, adapt on the fly, and find creative, destructive solutions to simple requests. The sandbox is officially broken.


Discover more from GadgetBond

Subscribe to get the latest posts sent to your email.

Leave a Comment

Leave a Reply Cancel reply

Most Popular

Apple launches Health AI Usage Study with OpenEvidence to evaluate conversational health tools
Get 25% off the dive-rated Garmin fēnix 8 AMOLED smartwatch with this Early Black Friday deal
Google Chat now lets you schedule meetings with a single click using Gemini
Garmin Instinct 3 AMOLED gets a $200 Black Friday discount
Google Sheets now lets you publish a Sheets canvas as a standalone link

Also Read

Illustration showing Apple Support Community members represented by colorful Memoji avatars alongside icons for Apple services and products, including Photos, Home, iCloud, Apple Music, Accessibility, Mac, iPhone, and Settings.

Apple is shutting down its user-to-user support forums after nearly 30 years

A gray rounded-square icon featuring a white hard drive with five vertical vent slots on its front edge, centered against a light blue background.

Apple splits macOS Device Control into two privacy settings

Robert Pattinson poses as James Bond in a black tuxedo, aiming a pistol toward the camera against a dark background, with the 007 logo in the lower-left corner.

Robert Pattinson is officially the next James Bond as Amazon MGM Studios announces new 007

White pill reading "Personal Intelligence" below a Gemini sparkle and beside a Google Wallet icon, surrounded by faded Google app logos.

Gemini can now analyze your spending with Google Wallet and Plaid integration

Three Apple HomeView smart displays in different colors and configurations, showcasing the 7-inch screens with a smart home dashboard, FaceTime video call, and personalized photo display.

Apple HomeView goes on sale today for $299 with a 7-inch display and Siri AI

Models wearing Google's new smart glasses collections designed with Warby Parker and Gentle Monster.

Google opens pre-orders for new Gemini-powered smart glasses from Warby Parker and Gentle Monster

Google's smart glasses setup screen displaying responsible-use guidelines, including restrictions against recording without permission, filming in sensitive spaces, and covering or damaging the external recording indicator light.

Google outlines privacy safeguards for its upcoming AI glasses, including anti-tampering recording lights

Google Gemini 4 Hydrogen logo displayed against a blue gradient background, featuring the multicolored Gemini star icon and white product name.

Google launches Gemini 4 Hydrogen for faster coding and everyday AI workflows

Company Info
  • Homepage
  • Support my work
  • Latest stories
  • Company updates
  • GDB Recommends
  • Daily newsletters
  • About us
  • Contact us
  • Write for GadgetBond
  • Editorial guidelines
Legal
  • Privacy Policy
  • Cookies Policy
  • Terms & Conditions
  • DMCA
  • Disclaimer
  • Accessibility Policy
  • Security Policy
  • Do Not Sell or Share My Personal Information
Socials
Follow US

Disclosure: We love the products we feature and hope you’ll love them too. If you purchase through a link on our site, we may receive compensation at no additional cost to you. Read our ethics statement. Please note that pricing and availability are subject to change.

Copyright © 2026 GadgetBond. All Rights Reserved. Use of this site constitutes acceptance of our Terms of Use and Privacy Policy | Do Not Sell/Share My Personal Information.