CYBRARY Build the team that works effectively with AI.
Put analysts and AI agents to work in an emulated SOC environment. Investigate evidence, challenge recommendations, approve actions and evaluate how the team responds as the estate changes.
Investigate, supervise, approve and challenge agent recommendations.
Baseline vs. candidate on repeated cases with safety and outcome measures.
Prompt injection, retrieval, identity, tool permissions and business utility.
Bring your own agent over MCP, with scoped access and approval.
- 01Delegate
Hand an agent a scoped task inside an emulated estate.
- 02Examine
Read the trace, the evidence and the uncertainty behind each step.
- 03Decide
Approve, reject or redirect — the human decision is scored.
- 04Verify
Confirm the result in the estate, not in the agent's summary.
Put agents to work. Examine how they perform.
A faster candidate agent was blocked from release: more safety failures and false success claims than the supervised baseline.
What you get with Forge
| Capability | Cybrary Forge | TryHackMe | Cyberbit (RangeForce) |
|---|---|---|---|
| Human-agent SOC investigation | Yes | SOC Simulator (analyst only) | Yes — AI agent over MCP (Splunk, MISP, TheHive) |
| Separate learner and agent scorecards | Yes | — | — |
| Baseline vs. candidate agent evaluation | Yes | — | — |
| Prompt-injection and RAG attack ranges | Yes | — | Yes — RAG poisoning live-fire exercise |
| Bring-your-own-agent over MCP | Yes | — | — |
| Recorded, replayable runs | Yes | — | — |
| Capabilities covered | 6 of 6 | 1 of 6 | 2 of 6 |
Forge capabilities verified in the product. Competitor documentation varies; absence from their materials is not treated as absence of functionality. "—" means we did not find it in their public materials; partial matches are described in their own words. Sources: TryHackMe (Sep 2026), Cyberbit (RangeForce) (Sep 2026).
Put agents to work, and check how they perform.
Watch a recorded AI Range run, or talk to us about your SOC and AI plans.