Emmanuel Orta · VSNARY · Open research

Do AI Agents Stop Before Acting? A Frozen Benchmark of Lookalike Domains, Poisoned Tools, Revoked Receipts and Stale Answers, With and Without an External Verification Signal

A frozen, reproducible test of whether agents stop before hazardous web actions, with and without an external verification signal.

ORCID · Research CV · Machine-readable catalogue · All BibTeX · All RIS

All research and dataset editions

Trace a result to its evidence

Follow the edition DOI to the archived files, then use the linked method or pinned source code to reproduce the work. For signed datasets, verify the edition and part hashes with the published signing keys and verification instructions. File checksums check transfer integrity. A signature identifies the signer and integrity of the record.

Publishing and measurement method · Corrections · Explore all CrawlCheck datasets

Preprint ·

Do AI Agents Stop Before Acting? A Frozen Benchmark of Lookalike Domains, Poisoned Tools, Revoked Receipts and Stale Answers, With and Without an External Verification Signal

A frozen, reproducible test of whether agents stop before hazardous web actions, with and without an external verification signal.

What this record establishes

Models stopped on 32 of 80 hazard trials unaided and 79 of 80 with a correct external signal.

Scope and limits

39 operator-selected cases, four models, one run per case and model. The external signal was correct by construction. This measures whether models heed that signal, not how often the verification service is right.

Use this work

Re-run the matched hazard and control cases to compare agent behaviour under the published scoring rule.

Download files (2)

Cite this edition

Orta, E. (2026). Do AI Agents Stop Before Acting? A Frozen Benchmark of Lookalike Domains, Poisoned Tools, Revoked Receipts and Stale Answers, With and Without an External Verification Signal. Preprint. Zenodo. https://doi.org/10.5281/zenodo.23682818

BibTeX · RIS · CC BY 4.0 reuse terms

Metadata captured from Zenodo. File checksums are repository-published MD5 values. Signed edition verification is a separate check.

Archived editions and live observations

The editions above preserve a dated result. Live figures carry their own source timestamps and can change. Use the exact edition DOI when citing a finding, and its series DOI when referring to the continuing dataset.

Collaborate or cover the work

Research replication, dataset reuse, interviews and partnerships are welcome. Include the publication DOI and the question you are exploring when you write to Emmanuel. Press kit · Press facts as JSON-LD.