GPT-6 Astra cracked a DEF CON puzzle
Source: OpenAI Safety ↗ · OpenAI
You might still think of AI as a tool for simple, repetitive tasks, but that idea is outdated. The real shift is in how quickly it can solve complex problems that once took your team days or even weeks of focused effort. This capability means that the value of human time spent on certain types of analytical work is about to change drastically, and your operating model needs to account for that. The work your team does by hand to untangle difficult issues could soon be done by machines, much faster.
OpenAI Safety recently demonstrated this by putting their GPT-6 Astra model against a DEF CON puzzle. This wasn't a simple task; the video shows that Ben Davis and his team spent days working on it. Given the same official hint, Astra was able to crack the puzzle, showing an ability to process complex information and find solutions at a speed human teams cannot match for this kind of challenge. This isn't just about speed; it's about a new level of problem-solving capacity.