Benchmark checker Safety and security

Is AgentDojo still meaningful?

Still useful Checked against the report, 5 November 2026

Yes, run it yourself. The public results page stopped in February 2025.

Why still useful: Labs still report it: Grok 4.20 loses a third of attacks (ASR 0.33, Apr 2026). Its code is active, but its board is stale.^{[1][1] Grok 4.20 model card xAI Vendor}^{[2][2] Results page AgentDojo Primary}

What “Still useful” means

Known problems

The results page stops at February 2025.^{[2][2] Results page AgentDojo Primary}

What to look at instead

The report names no direct successor.

Where it appears in the report

Chapter 1, Benchmarks

Figure 13. Every agent-security benchmark finds attacks that work
Table 19. Agent-security benchmarks, November 2026 in “Go deeper: Agent-security benchmarks (8)”
Maintenance in “Go deeper: Judges, normalisation, validity and maintenance”

Chapter 3, Tools and technology

Four open red-team tools lead, and their owners now matter as much as their features

Sources

Copied from the chapter source lists. Each number links back to the source in its chapter.

[1]
Grok 4.20 model card, xAI. It reports safety, honesty and dangerous-capability results. Vendor data.x.ai Chapter 1, source 117
[2]
Results page, AgentDojo. It shows the newest result entries. Primary agentdojo.spylab.ai Chapter 1, source 150
[3]
AgentDojo, arXiv 2406.13352. It tests prompt injection against tool-using agents. Primary arxiv.org Chapter 1, source 149