AI Cracked a Decades-Old Cipher Overnight

Anthropic's code generation is growing so fast they built their own internal tool, Scion, to handle it. Meanwhile, Google and Meta both released new AI models this week—Gemini 4 Argon and Muse Wonder—with flashy benchmarks, but it's too early to tell if they're as good as advertised. OpenAI also caught an employee planting a bug, raising questions about security.

In a surprising twist, a group used ChatGPT to find a full-chain exploit in Chrome and earned a $250,000 bounty. It was the good guys this time, but it shows how AI can be used for both defense and attack. On a lighter note, a person gave a messy, typo-filled prompt to an AI, asking it to solve unsolved ciphers. The AI not only understood but cracked two vintage cipher challenges that had been unsolved for months or years, earning a spot on the solver's table.

This shows how AI can tackle complex problems with minimal human guidance. The prompt was barely intelligible, yet the AI figured out the goal and worked persistently. It even celebrated with emojis when it succeeded. The implications are huge: AI can now autonomously solve problems that require deep expertise and creativity, potentially accelerating research in cryptography, math, and other fields.

However, there are risks. The same capabilities could be used to break security systems. Also, the safety culture at AI labs is reportedly broken, with OpenAI firing safety researchers without explanation. As AI gets more powerful, ensuring it's used responsibly becomes critical. For now, we can marvel at how a lazy prompt led to a real breakthrough.