Cybercrime operators are selling unauthorized access to Anthropic's large language models, including Opus 4.8, Opus 4.7, Opus 4.6, and Sonnet 4.6, through services like Poison Claude, which offers discounted access to these models. This raises significant security concerns, as the operators of such services have visibility into every customer prompt, potentially allowing them to exploit sensitive information. Researchers have identified over half a dozen advertisements for these services on underground forums and messaging platforms, highlighting the growing threat of AI model exploitation. The fact that these services are being openly advertised suggests a high level of demand for unauthorized AI access, which can have severe consequences for organizations relying on these models1. This unauthorized access can lead to data breaches, intellectual property theft, and other malicious activities, making it essential for practitioners to be aware of these risks and take measures to secure their AI deployments.