Hack The Box Scores Agents; AI Range to Weigh

Hack The Box adds an Enterprise Edition

SiliconANGLE reported on October 6, 2026, that Hack The Box Ltd. introduced AI Range Enterprise Edition so security teams can test whether their own AI agents do the jobs they were assigned. Duncan Riley wrote the report, which describes Hack The Box as a cybersecurity training company.

Business Wire carried the company's own announcement the same day.

Help Net Security said on October 6, 2026, that Hack The Box made its platform for testing and measuring AI security agent effectiveness available to enterprise security teams.

Tech Insider reported on October 8, 2026, that AI Range itself isn't brand new and that the Enterprise Edition is what launched this week. Tech Insider added that the move pushes Hack The Box, long known for Capture The Flag challenges and certification tracks for human penetration testers, into certifying whether AI agents can do the work companies assign them.

Agents are scored against assigned roles

Tech Insider said the Enterprise Edition assigns an agent a defined cybersecurity role and evaluates it under realistic, pressure-tested conditions instead of grading it against an abstract leaderboard.

Hack The Box says an agent may be put into production after a strong benchmark result or vendor test without being checked against the work it will handle day to day, SiliconANGLE reported. SiliconANGLE also reported that an agent's performance can drift after deployment as the model underneath it gets updated.

The appraisal method is called Agentic Worker Competence, SiliconANGLE said. Each agent gets a score for its assigned role and a pass-or-fail mark in each environment it works through, SiliconANGLE reported. New environments are added regularly as fresh vulnerabilities and attack patterns turn up in the wild, SiliconANGLE said.

MSSP Alert said on October 7, 2026, that customers can rerun assessments whenever an agent, its underlying model or its data is updated. MSSP Alert credited SiliconANGLE for the underlying details.

Hack The Box said organizations can build "recurring, evidence-based proof" that an agent can perform a specific job role, Help Net Security reported.

Human staff are graded next to agents

SiliconANGLE said in its October 6, 2026, report that penetration tester and security operations center analyst are the only two roles with AI-augmented versions at launch.

In the analyst version, practitioners lean on AI to review alerts in realistic scenarios, and whatever the AI recommends doing next is theirs to evaluate, SiliconANGLE said. The scenarios test whether a practitioner catches errors in an agent's reasoning and knows when to step in. Holding back is scored as well, based on whether someone lets AI-driven work run when it does not need correcting.

Help Net Security said Agentic Operator Competence Scoring evaluates whether professionals can question AI agents' work and intervene. Hack The Box calls the system proprietary and industry-first, Help Net Security reported. The outlet added that additional cybersecurity roles are planned in the coming months.

Haris Pylarinos, founder and Chief Executive of Hack The Box, said leaders need "confidence in both sides of the workforce" as AI takes on more cybersecurity tasks, SiliconANGLE reported.

AI Range began in December 2025

Hack The Box introduced the original AI Range in December 2025, Help Net Security said in its October 6, 2026, report.

AI labs, government customers and hyperscalers have pressure-tested the platform since, according to the company, SiliconANGLE reported.

The Hack The Box website, dated October 6, 2026, describes its first AI Range as "the world's first controlled AI cyber range built to test and benchmark the safety, limits and capabilities of autonomous AI security agents."

Datacentre reported on October 6, 2026, that organisations can test their own AI security agents and measure how well staff work alongside them. App Dealroom carried the launch on October 7, 2026.

Tech Insider wrote that as more firms hand off tasks like triage, log review and exploit development to autonomous agents, the gap between a strong benchmark score and a reliable agent has become one of the more pressing questions in enterprise cybersecurity.

Tron's take

I read the news as a launch aimed at enterprise security teams, which is how Help Net Security and SiliconANGLE describe the audience. I would not plan around buying it. That is my reading of the news, not a reported result.

The idea inside it travels better than the product. SiliconANGLE and Help Net Security both reported that an agent's performance can change as models, software and data change. Any owner can put that point to a vendor selling an agent: ask what the agent was tested on and when it was last retested. This site holds that most small and mid-sized businesses do better applying proven capabilities well than chasing each release. I would file this launch under things to know about, not things to act on this week.

Earlier XL.net coverage of SOCs facing human pressure as AI speeds alerts and of benchmark limits shown by Kimi K3 covers adjacent ground.

XL.net sells managed IT and security assessments. A business weighing an AI agent for security work could ask XL.net or any other provider to review how the agent is checked, and I would want that review to cover retesting after model changes.