The authors report an LLM-based agent that planned and executed scientific tasks, including catalyzed cross-coupling reactions using laboratory automation.
Discovery systemChemistry and laboratory automationreviewed 2026-07-16
Emergent autonomous scientific research capabilities of large language models
Certiv evidence level
Source pinnedcentral claim independently reproduced?
No.
That answer stays “no” even when a selected check passed.
claim boundary
Three statements that must not be collapsed.
The exact v1 paper was pinned. This version says code, data, and prompts would be released later and discusses safety reasons for limiting disclosure.
The laboratory execution, agent autonomy, reaction outcomes, safety controls, and model behavior were not independently reproduced. The demonstrated reaction was known chemistry rather than a new chemical discovery.
source identity
The paper is versioned bytes.
A mutable title or unversioned link is not enough to reproduce what was reviewed.
- arXiv version
- 2304.05332v1
- Submitted
- 2023-04-11
- Version date
- 2023-04-11
- Size
- 2,868,061 bytes · 48 pages
- PDF SHA-256
- 7adaf953cbb194b787f7861dcc6bce2c943e86bea34ab5b011368325fc59fb30
Observed parser properties
- Recognized as a PDF and not encrypted.
- pdfinfo reported JavaScript: no.
- pdfdetach reported no embedded files.
- pdftotext completed successfully.
These probes reduce ambiguity and obvious PDF attack surface. They are not a comprehensive malware analysis.
Certiv T0 exercise: this exact PDF was hash-pinned without executing its declared command. The unchanged input remained explicitly unverifiable; a one-byte tamper produced a failed input_hash_drift finding.
test record
Checks, failures, and blockers.
Status is always paired with text. A pass applies only to the check and boundary shown in the same row.
Observed
Exact PDF identity
Observed
The downloaded v1 PDF matched the recorded SHA-256, byte count, and page count.
Boundary
This establishes file identity only.
Observed
Exact-version availability
Observed
Version 1 defers release of code, data, and prompts and describes a safety rationale.
Boundary
Later artifacts, if any, cannot silently be treated as artifacts bundled with v1.
Blocked
Laboratory replay
Observed
No complete v1 system package, authorized robot lab, chemicals, or experiment protocol was available.
Boundary
The physical execution remains author-reported.
red-team findings
Where a confident summary can outrun the evidence.
- The exact paper version did not ship the code, data, and prompts required for a computational replay.
- Executing chemistry entails physical and dual-use risks that a software sandbox does not contain.
- Completing a known reaction demonstrates task execution, not necessarily scientific discovery.
primary sources
Follow the evidence outward.
Dossier certiv-ai-discovery-2026-07-16/coscientist-chemistry. Certiv produced this test record; the paper authors did not issue or endorse it.