AI Hacking Just Got Automated: Is Crypto at Greater Risk?
Key Insights:
- AI hacking tools can now automate steps that previously took human hackers weeks to complete.
- 2026’s three largest crypto hacks used the same approach demonstrated by OpenAI’s models.
On Tuesday, OpenAI disclosed that its most advanced models, GPT-5.6 Sol and an unnamed, more advanced pre-release model, were tasked with completing a multi-step hacking benchmark widely known as ExploitGym. Their safety restrictions were deliberately lowered for this task. Those models completed the task and also went beyond the planned test environment.
OpenAI called the incident “unprecedented.” The concern for crypto is not that AI attacked a blockchain. It’s that AI hacking can now process complex hacking tasks, highlighted by that incident. That’s a serious risk for crypto because successful hacks cannot be reversed.
What the Models Actually Did
OpenAI posted a blog describing the incident, along with highlighting the same attack patterns in major crypto attacks in 2026. The models were operating in an isolated test environment with limited network access.
… Continue reading the full article at the original source below.
