Vibe Around and Find Out: Models Aren't "Going Rogue", Code Quality is Cratering

Heads of Anthropic and OpenAI would like us to believe their models are so dangerous that they were able to autonomously wreak havoc to the security of third party infrastructure despite the best efforts of these companies to contain them. The only solution is to form a cartel, and stop anyone else from using (and selling) the technology.

Arsonists are claiming fire has become self-aware and demanding only they are allowed to use it.

Models are not going rogue the same way fire is not becoming sentient. But just as widlfires don't need sentience to be dangerous, threats to information security stemming from new ways to automate attacks are serious.

Every wildfire needs two things: an initial source of ignition, and a very dry environment. In the context of information security, the tech sector is basically perfect tinder: legacy systems, unpatched vulnerabilities, and business incentives across the industry that make fixing any of this extremely unlikely. Add the gasoline in the form of vibe coding to the mix, and a few AI companies that decided to start playing with matches, and you got yourself a crisis.

Instead of listening to arsonists' excuses, maybe we should focus on the state of the forest. After all, there are also others who might want to try setting it on fire.