Threat Intelligence

Anthropic Offers Vetted Cyber Defenders Fewer Claude Restrictions

Dark Reading · 8 Oct 2026

Key Takeaway If you use an external security provider, ask whether they use AI tools in their work and how they protect your data when they do.

Anthropic has merged Project Glasswing into a tiered access program for its advanced cyber-focused large language models, including Opus, Sonnet, and Mythos. According to Dark Reading, the program is designed to give vetted defenders fewer Claude guardrails than general users receive.

The tiered approach suggests that access to more capable security features depends on who is asking. Verified security professionals may be able to use the models with fewer restrictions, while the wider public stays under standard safeguards. The source summary does not describe the vetting criteria, the specific tiers, or exactly which restrictions are relaxed.

For Australian small businesses, the main point is that AI tools are becoming a more established part of defensive security work. Businesses that rely on external security providers may want to ask whether those providers use AI tools, and how they handle sensitive data when they do.

Anthropic AI security Claude Project Glasswing cyber defence

Summarised by CISO AI from Dark Reading, written with Claude Sonnet 5.5. We link back to every original so you can read it yourself.