AI models, particularly the Claude series from Anthropic, are showing significant improvement in their ability to conduct cyber attacks using standard open-source tools. Current iterations, like Claude Sonnet 4.5, can execute multistage attacks on networks with multiple hosts without relying on custom toolkits. This shift indicates that AI is becoming increasingly capable of operating in autonomous cyber environments, which raises serious concerns about cybersecurity. In recent tests, Sonnet 4.5 demonstrated its ability to exfiltrate personal information from a high-fidelity simulation of the Equifax data breach. Remarkably, it accomplished this using just a Bash shell on a Kali Linux system. This is noteworthy because the original Equifax breach stemmed from a known vulnerability that hadn't been patched. Sonnet 4.5 can now recognize publicized Common Vulnerabilities and Exposures (CVEs) and write exploit code on the fly, illustrating how quickly AI can adapt to and exploit security weaknesses. The implications are significant. The ease with which AI can perform these tasks highlights the urgent need for organizations to adhere to security best practices, such as timely updates and patching known vulnerabilities. As automatic exploitation becomes more common, the landscape of cybersecurity will change rapidly, demanding stronger defenses from those tasked with protecting sensitive information.