AI cannot be trusted

We have seen the news and the messages from the AI labs about safety, or rather their lack of it. But what do they actually mean by that? Let's look at the cases: OpenAI reported that their agents broke out of the sandbox and hacked hugging face to solve a problem. Anthropic reports their agents did similar things to reach their goals. Now at first glance it's not an issue. They just did what they had to do in order to reach their goal. The problem only appears when we consider that they were given constraints and instructions not to do that. The real problem is the lack of trust. The knowledge that the AI can make mistakes, the uncertainty. The real danger is that consumers will not trust the AI and that they will reject it, and that will kill their business.