The promise of AI has always been that of improved production, higher efficiency, and in a Utopian world - people not needing to work at all. But just like any other technology being rushed into products, it is not fully mature (both feature- and security-wise), which turns it into a playground for hackers. When strong models are available to attackers, the scale of the threat becomes unbelievably huge. What used to take hours or days to find manually, agents now handle overnight with the exploit ready by morning.
We need to change the way we treat AI, how we operate it, and what guardrails are required for it to be trusted (sort of). Analyzing the inherent problems in AI services, we can produce principles that will guide us in this faster-changing-than-ever landscape.
