Anthropic Widens Vetted Cyber Access and Launches a Defense Mission

Anthropic made two connected cybersecurity announcements this week. On October 6 it expanded its Cyber Verification Program into three application tiers. The first is aimed at defensive teams; the second covers authorized red teams; the most specialized tier is for organizations testing systems whose failure could affect lives or markets. The program offers qualified applicants access to advanced models with fewer cyber restrictions, while retaining verification and monitoring requirements. On October 8 Anthropic launched a Cyber Mission, with a critical-infrastructure program for selected partners and a free scanner for open-source projects. The company says engineers and threat research will accompany its models in some of that work. These steps acknowledge a difficult balance: defenders need stronger tools, while the same capabilities could help attackers. Access is therefore selective, not a general removal of safeguards. Anthropic reports promising evaluation results, but those are its own tests. It also concedes that finding weaknesses is easier than verifying, prioritizing and fixing them. The practical measure will be whether participating teams actually reduce risk in power, water, transport and widely used software.