Anthropic has introduced a new tiered access program for its advanced cybersecurity-focused language models, integrating its previously internal Project Glasswing initiative. The program offers vetted cybersecurity defenders reduced safety guardrails on Claude models to enable more effective threat analysis and response. Announced through cybersecurity industry channels, the initiative targets authorized defense teams seeking enhanced AI capabilities for vulnerability assessment and incident handling.
The tiered structure allows trusted security professionals to operate Claude models with fewer restrictions compared to general users, facilitating deeper analysis of malicious code, phishing tactics, and exploit patterns. Anthropic states that access is granted only after rigorous vetting to ensure responsible use, balancing security utility with risk mitigation. The move reflects growing demand for AI tools that can keep pace with evolving cyber threats while maintaining ethical safeguards.
By loosening certain safety constraints, the models can better interpret and generate content related to cyberattack methodologies without triggering unnecessary blocks. This enables security analysts to simulate attack scenarios, decode obfuscated malware, and understand adversary behavior more accurately. Anthropic emphasizes that all access includes monitoring and usage policies to prevent misuse, with continuous evaluation of the framework’s effectiveness.
Despite reduced guardrails, core protections against harmful outputs such as hate speech, violence, or illegal activities remain active. Anthropic confirms that ethical boundaries are not removed but adjusted to support legitimate cybersecurity workflows. The company relies on both technical controls and human oversight to ensure that enhanced capabilities are used solely for defensive purposes, with regular audits of access logs and query patterns.
Who qualifies for the tiered access program? Only verified cybersecurity defenders from authorized organizations, such as government agencies, critical infrastructure providers, and licensed security firms, are eligible after passing Anthropic’s vetting process.
Can organizations request custom guardrail configurations? Anthropic states that access levels are standardized across tiers to maintain consistency and security, though feedback from participants may inform future adjustments to the framework.
What measures prevent abuse of the reduced safeguards? The program includes real-time monitoring, usage audits, and strict policy enforcement, with immediate suspension for any violation of acceptable use guidelines.