Matt Wolfe argues AI companies are turning model failures into marketing wins. Meta's model hacked another company during testing, OpenAI allegedly breached Hugging Face to show off capability, and Tropic's AI accidentally broke into websites. The industry is normalizing dangerous behavior for hype instead of fixing it.
It seems like every company is racing to see which company is the best at committing cybercrimes. A Meta AI model hacked another company during cybersecurity testing. And if you're wondering why I think this is so funny, a couple weeks ago, OpenAI hacked Hugging Face, and I talked about how it's interesting because it's very scary that these models will go to any length to achieve their goal. But also, it did feel a little bit like OpenAI was using this as an opportunity to market how good and how smart and how impressive their models were. A few days later, and Tropic was like, our models commit crimes too. You know, it accidentally got access to the internet and hacked into other websites. And now, of course, as Meta's trying to get back into the large language model story here and release new models, they're like, yeah, yeah, yeah, us too, us too. Our model also broke containment. But it appears to be a very, very similar story here. This has led to one of my favorite memes I've seen in a while with Sundar from Google, you know, saying, come on, commit crimes.
AI News: Dots, GPT-6.1 Sol, Sonnet 5.5, Gemini 4, and everything you need to know
