The authors report that AI-Newton derived general physical laws from raw multi-experiment data without prior physical knowledge, rediscovering several established laws.
Discovery systemPhysicsreviewed 2026-07-16
AI-Newton: A Concept-Driven Physical Law Discovery System without Prior Physical Knowledge
Certiv evidence level
Artifact inventoriedcentral 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 v2 PDF and repository revision were pinned. The documented stack requires Python, Rust, CUDA, and Maple 2024; Maple was not present in the review environment.
The benchmark, days-long full run, absence of prior knowledge, robustness to noise, and general discovery capability were not independently reproduced. The reported outputs are rediscoveries of known laws, not newly established laws of nature.
source identity
The paper is versioned bytes.
A mutable title or unversioned link is not enough to reproduce what was reviewed.
- arXiv version
- 2504.01538v2
- Submitted
- 2025-04-02
- Version date
- 2025-12-11
- Size
- 5,198,153 bytes · 6 pages
- PDF SHA-256
- bce6758e3ec33fe715b2df7132733107f98f1e928abb9ead89b9b7e2b3a8cdc1
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 v2 PDF matched the recorded SHA-256, byte count, and page count.
Boundary
This establishes file identity only.
Observed
Repository and dependency inventory
Observed
The repository was pinned at c143e865be42e067faf64cd8117cbeffd0fccfb6; its README, Cargo.lock, and sample knowledge file were separately hashed.
Boundary
Inventory records what was reviewed, not whether the system works as claimed.
Blocked
Full benchmark replay
Observed
Maple 2024 was unavailable, and the documented full test requires several days on high-end servers.
Boundary
The missing proprietary dependency and compute budget prevented a faithful run.
external artifacts
Pinned, not trusted by default.
AI-Newton repository
Observed revisionc143e865be42e067faf64cd8117cbeffd0fccfb6
README SHA-256 e35a6fd49f3dfa9600b400d906ef46146e3ec9d0c179c985a313b89fbfe2a4ab; Cargo.lock SHA-256 1e75c6930405e1ec20be9936c6ca9d75e8b42f3d8a61462d33a18d244cf8920c; sample knowledge SHA-256 a7524f42a48dc6165eabd65c9fc3696fdd652df8dc1e4d251bdc93ebdb956df9.
red-team findings
Where a confident summary can outrun the evidence.
- A proprietary Maple dependency prevents a fully open replay even though the repository is public.
- The short benchmark and full multi-day benchmark have materially different resource requirements.
- Rediscovering known laws is evidence about a system's behavior, not a new physical discovery.
primary sources
Follow the evidence outward.
Dossier certiv-ai-discovery-2026-07-16/ai-newton. Certiv produced this test record; the paper authors did not issue or endorse it.