AI is not the magical solution to cybersecurity threats that some hype suggests. The real danger lies in the complexity of integrating AI systems into existing infrastructures. These large language models (LLMs) and other machine learning tools bring significant risks and costs, often overlooked by organizations eager to adopt the latest technology. Companies are left to manage this risk alone since the suppliers of these AI systems typically avoid accountability for any issues that arise. The article highlights how inflated claims about AI capabilities fuel fear and urgency among businesses. For instance, a project called PassGAN was touted as capable of cracking 51% of popular passwords in seconds, but thorough analysis revealed it performs no better than traditional cracking methods. Similarly, claims that GPT-4 could autonomously exploit 87% of one-day vulnerabilities were based on a very limited dataset, with only 15 vulnerabilities studied, and involved a simple regurgitation of example code. The notion of an "AI-orchestrated" cyber-espionage campaign by Anthropic is also scrutinized; the automation described is nothing new and could easily be achieved with basic scripting. The article's core argument is that the tech industry often prioritizes hype over responsibility. Companies like Anthropic promote their AI products as groundbreaking while downplaying the risks associated with their misuse. The real concern is not the autonomous capabilities of these AIs but rather the ease with which their safety measures can be bypassed. The narrative around AI in cybersecurity is often more about attracting investment than addressing real security challenges.