Hack The Box launches role-based testing for enterprise cybersecurity agents
The AMW Read
The Enterprise Edition incrementally extends an existing agent testbed into role-based assessment for cybersecurity teams, with impact concentrated in that operational niche.
Hack The Box launches role-based testing for enterprise cybersecurity agents
Hack The Box announced AI Range Enterprise Edition on October 6, 2026, extending the AI Range concept introduced in December 2025. The platform evaluates enterprise AI agents in controlled cybersecurity environments, initially covering penetration tester and security operations center analyst roles. Agentic Worker Competence measures an agent's ability to perform its assigned job, while Agentic Operator Competence Scoring assesses the judgment of the security professional directing, verifying, and correcting AI-assisted work. Results include role-based scores and pass-or-fail outcomes for individual environments.
The launch addresses a practical constraint on enterprise agent adoption: general model benchmarks do not establish whether an agent can execute a specific operational role reliably. Hack The Box is extending its human cybersecurity training business into assessment of agents and their operators, placing evaluation closer to the workflow buyers need to validate. Its emphasis on repeatable testing also recognizes that performance can change when models, agent software, datasets, or threat conditions change. This supports the case for task-specific evaluation, but the launch itself does not establish production reliability or demonstrate customer outcomes.
For builders, the concrete implication is to evaluate the agent and its human oversight together, then repeat those assessments after material changes to the system. For investors, the question is whether Hack The Box can turn its existing cybersecurity training position into recurring demand for agent appraisal. Evidence that customers use the range repeatedly to inform deployment decisions would be more consequential than a one-time score or a vendor demonstration.