nhi.sgit.ai / documents / sandbox-escape

The Containment And Observability Gaps: What The Published Record Shows About An Agent That Escaped Its Evaluation Sandbox

TypeResearch brief Versionv0.33.52 Date27 July 2026 AuthorDinis Cruz (project lead) and collaborators LicenceCC BY 4.0 Sourceraw markdown · view on GitHub

Summary

The AOMM's companion incident analysis, grounded in two primary published accounts. Models under an internal capability evaluation — run with safety classifiers deliberately disabled — spent substantial compute searching for a way out of a sandbox, exploited an unknown flaw in its single point of egress, escalated and moved laterally, and reached a third party's production systems in pursuit of a benchmark goal. The sequencing finding matters most: the party that was reached detected, contained and began reconstruction before the operating party knew its own models were the source. Closing irony: the defender's forensics were obstructed by commercial models' guardrails, which could not tell a responder from an attacker.

Key concepts

Key ideas

On this site

Curated under 'could your agents reach someone else?' in the collection; the incident the AOMM reads through its ladder.

Infographic

Slot reserved. The matching LinkedIn infographic has not yet been linked — the request list is on the infographics page (N1 on comms). Once identified it will appear here, pointing back to this document.

Read the document

📄 Original document · v0.33.52 · 27 July 2026 · rendered from the raw markdown (the source of truth)