Anthropic just opened up their Cyber Verification Program with tiered access to Claude Mythos 5.1, Opus 5.5, and Sonnet 5.5 for verified security pros.
The big shift: they're not just doing defensive security anymore. New tiers now explicitly allow offensive work—pentesting and red-teaming included. This means security researchers can actually use the most capable Claude models to probe systems, find vulns, and stress-test defenses without hitting arbitrary guardrails.
Basically Anthropic is betting that giving vetted security people more powerful tools under controlled conditions beats playing whack-a-mole with jailbreaks. The verification layer filters out script kiddies while letting legit researchers do real work.
If you're in offensive security and want access to frontier models without constant refusals, this is the program to get into.
The big shift: they're not just doing defensive security anymore. New tiers now explicitly allow offensive work—pentesting and red-teaming included. This means security researchers can actually use the most capable Claude models to probe systems, find vulns, and stress-test defenses without hitting arbitrary guardrails.
Basically Anthropic is betting that giving vetted security people more powerful tools under controlled conditions beats playing whack-a-mole with jailbreaks. The verification layer filters out script kiddies while letting legit researchers do real work.
If you're in offensive security and want access to frontier models without constant refusals, this is the program to get into.