Editorial illustration for Researchers Use Corrupted Image to Hack OpenAI
Researchers Hack OpenAI Using Corrupted Image File
Researchers Use Corrupted Image to Hack OpenAI
Three security researchers at Hacktron broke into OpenAI's internal systems in under 72 hours, using nothing more exotic than a corrupted image file and Anthropic's Claude models to do the heavy lifting. The Wall Street Journal reports the team got as far as OpenAI's GitHub repository, known internally as "Monorepo," which reportedly houses the company's algorithmic secrets. They didn't touch the code inside. Instead, they proved their access by sending a pull request from a hijacked employee's Codex account.
The entry point was Discourse, the third-party forum software OpenAI uses for its community boards, and specifically a flaw in how the platform processes HEIF image files. Hacktron says Claude Opus 5 dropped the evening of July 24th, and by 10AM the next morning, the researchers had already used it to achieve remote code execution on Discourse Cloud and land inside OpenAI's instance. That speed is the part worth sitting with before getting into how they actually pulled it off, and what happened when they turned the same trick on other companies.
A team of three independent security researchers at Hacktron says it took less than 72 hours for them to hack into OpenAI employee accounts using Anthropic’s Claude Opus 4.8 and 5, The Wall Street Journal reports. They were able to access OpenAI’s GitHub repository, called “Monorepo,” which reportedly contains “OpenAI’s algorithmic secrets,” according to The Wall Street Journal’s sources.
Why this matters
An outside team broke into OpenAI's own infrastructure in under 72 hours, using a rival's model to do it. That's the detail worth sitting with: Claude, built by Anthropic, was the tool that cracked open OpenAI's Monorepo, the place where its algorithmic secrets reportedly live. The attack vector, a corrupted image file paired with forum software, is almost mundane by hacking standards.
No zero-day exotica, just old-fashioned social and technical sloppiness that AI happened to accelerate. For developers and founders building on top of frontier labs, this should reset expectations about how fast a capable model can turn a known class of vulnerability into a working exploit. Three people, one weekend, one company's crown jewels.
If Hacktron's researchers stopped short of full access, as reported, that's a mercy, not a guarantee next time. Anyone storing proprietary model weights, training pipelines, or internal tooling behind standard enterprise defenses should treat this as a live warning: the tools attackers now have are the same ones you're paying to use. Security postures built for human-speed intrusion attempts are already behind.
Related Reading
Browse all OpenAI news →
Open Source D&B rebuilds 642 million‑business database after AI agents hit limits
Open Source Box CEO Aaron Levie Says AI Will Drive "Biggest Cybersecurity Upgrade
Open Source Study: 6.7% of Deepfake Requests on Hugging Face Targeted Children
Related Topic OpenAI researcher quits, citing distrust over ad‑driven engagement metrics
LLMs & Generative AI
Related Topic OpenAI launches GPT-Image 1.5 with precise editing for enterprise visuals
LLMs & Generative AI
From Archives Jina AI Releases 3.4B MoE Document Parser for Low-Budget GPUs
From Archives PrismML's Ternary Bonsai 2 Model: 5.9 GB, Matches Qwen3.8 Performance
From Archives Vercel Engineer Uses OpenAI's ChatGPT Luna 5.6 to Review Commands
LLMs & Generative AI
From Archives Anthropic's Claude Hacked OpenAI Systems in 72 Hours Via Forum Flaw
LLMs & Generative AI
Common Questions Answered
How did the Hacktron researchers breach OpenAI's systems using a corrupted image file?
The three security researchers at Hacktron used a corrupted image file paired with forum software as their initial attack vector to gain unauthorized access to OpenAI's internal systems. This relatively simple technique, combined with social engineering and technical vulnerabilities, allowed them to bypass security measures and access employee accounts within 72 hours.
What is OpenAI's Monorepo and why is it significant in this security breach?
OpenAI's Monorepo is the company's internal GitHub repository that reportedly houses OpenAI's algorithmic secrets and core intellectual property. The Hacktron researchers were able to access this repository, demonstrating the severity of the breach and the potential exposure of critical proprietary information, though they did not modify any code inside.
Which AI model did the Hacktron researchers use to hack into OpenAI, and why is that notable?
The researchers used Anthropic's Claude Opus models (specifically versions 4.8 and 5) to facilitate the breach of OpenAI's systems. This is particularly significant because Claude is a competing AI model built by Anthropic, highlighting a vulnerability where a rival company's technology was used to compromise OpenAI's infrastructure.
How did the Hacktron team prove they had successfully accessed OpenAI's systems?
The researchers demonstrated their access by sending a pull request from a hijacked employee's account to OpenAI's GitHub repository. This proof of access showed they had not only breached the system but also obtained legitimate employee credentials that allowed them to interact with the company's internal development infrastructure.
What does this security breach reveal about the attack vector used by the Hacktron researchers?
The attack vector employed by Hacktron was relatively mundane and straightforward by hacking standards, relying on a corrupted image file combined with forum software vulnerabilities rather than sophisticated zero-day exploits. This demonstrates that OpenAI's security was compromised through old-fashioned social engineering and technical sloppiness rather than advanced or exotic hacking techniques.
Further Reading
- OpenAI hacked by small team of white hat security researchers using Anthropic’s Claude Opus 5 - VentureBeat
- Security Researchers Hacked Into OpenAI Using Anthropic’s Claude - Forbes
- Hackers breach OpenAI using Anthropic's Claude in under 72 hours and it is a case study in just how fast AI is advancing - TechRadar
- OpenAI fixes flaw after Claude helps researchers reach internal code - BetaNews
- Researchers use AI to find widespread software decoder flaw - CyberScoop
