— A loyal New York Times crossword fan after this travesty happened for the first time in the puzzle’s 84-year history. Meanwhile, we’re still stuck on one across.
Artificial Intelligence
Anthropic Might've Thrown Us Into an AI Cold War
What’s going on: AI can do more than summarize e-mails, create cartoon versions of your dog Millie, and displace workers. Certain programs could be well on their way into take-over-the-world territory, Brain-style. Anthropic announced this month that it built an AI model it considers “too dangerous” to release widely: Mythos. The company says it can exploit weaknesses in software that runs global banks, governments, and power plants. Naturally, world leaders are worried. There’s already a power imbalance in play — what began as about a dozen companies with access has expanded to 40 more. But it’s still a tight US-dominated circle (though a British government-backed AI security group has access). Mythos gives the US an unprecedented edge on rivals like Russia and China. One pro-Kremlin outlet compared it to a nuclear weapon. Meanwhile, the NSA may already be using the model (so much for that “supply chain risk” talk…).
What it means: Anthropic isn’t sharing this widely anytime soon and says it will coordinate next steps with the US government. But the clock is ticking faster than anticipated. An unauthorized group of coders accessed a preview version of Mythos due to surprisingly lax security, per a Bloomberg report. While this group is reportedly interested in “playing around” with new AI models rather than weaponizing them, it underscores the security risks as new models emerge. Anthropic expects similar cyber-capable AI to be out in the wild within 18 months. Silver lining? It could prompt some US companies to beef up their own security. One web browser company said Mythos found 271 vulnerabilities it can now fix.
