Anthropic Gives Vetted Defenders Fewer Claude Guardrails

Anthropic has officially integrated its Project Glasswing initiative into a new tiered access program designed for cybersecurity professionals. This update allows vetted security researchers and defenders to utilize the company's advanced large language models—including Claude 3 Opus, Sonnet, and the newly mentioned Mythos—with significantly reduced safety guardrails. By loosening these restrictions, Anthropic aims to empower experts to better identify vulnerabilities, analyze malicious code, and develop more robust defensive strategies without the friction typically imposed by standard safety filters. This move reflects a growing trend among AI developers to provide specialized access for the infosec community, acknowledging that the dual-use nature of generative AI requires a nuanced approach to safety. The program is expected to streamline threat intelligence workflows and improve the efficacy of AI-assisted security operations for organizations participating in the vetting process.