Adobe Content Authenticity
The AI review rates the findings as likely false positive, but the risk score (53/100) still counts them. The review covers version 1.0.2, not the analysed version 1.0.7.
Analysis record
- Analysed
- 4 days ago 18 November 2026
- Version
- v1.0.7
- Artifact
- SHA256 0cf3c375…df987c
- Detail
- Full list
Is Adobe Content Authenticity safe?
This is Adobe's browser tool for reading Content Credentials, the signed metadata that cameras and editing software attach to images and video. The heavy work runs in WebAssembly: c2pa_bg.wasm parses the credential manifests, and ort-wasm-simd-threaded.asyncify-9GUf3Unn.wasm is an open-source machine learning runtime used to analyse images on your machine. The extension declares no special permissions, and it requests no host permissions either, so it has no standing access to the pages you visit.
The network findings all sit in one bundled file, assets/imageProcessor.worker-BeCTT-Lm.js. If those calls were live, that worker would be shipping your images somewhere. What the scanner actually picked up is a list of strings such as ai.onnx.training and accountant.cc, which are ONNX Runtime operator names and C++ module fragments rather than real servers. Nothing in the code matched a malware signature. The two obfuscation hits, listed as OBFUSCATION-LARGE_WASM_FILE-c2pa_bg.wasm-0 and its ONNX counterpart, are size checks that fire on any large WebAssembly binary.
So why do the results look so loud? A scanner reading minified worker code and compiled WebAssembly turns up hundreds of strings shaped like domains and IPs. Hex fragments, dotted property chains and filenames all end up in that pile. Every network hit here traces back to a bundled file from Adobe's Content Credentials SDK and an open-source ML runtime, which is exactly the work the extension says it performs.
Evidence ledger
Ranked by severity · findings with a source location link to the code viewer
Publisher Evidence
Lowdev--d0173bdd-c13c-09c2-bcda-127cbd395e40
Publisher identity, store signals, distribution reach, and warning signals used for context. Treat this as supporting evidence, not a clean bill of health.
Chrome does not expose the same publisher verification data as IDE stores, so this score is deliberately conservative.
12 evidence rows available.
Finding Categories
Requested Permissions
5 permissionsAI Security Report
AI Security Review
Evidence context: threat category none; evidence quality moderate.
What this extension is
Adobe Content Authenticity (version 1.0.2, published under [email removed]) is the inspection half of Adobe's Content Credentials work. It reads the signed provenance manifests that cameras and editing tools attach to images and video. The heavy lifting happens in WebAssembly: c2pa_bg.wasm is the C2PA SDK compiled from Rust, and assets/ort-wasm-simd-threaded.asyncify-9GUf3Unn.wasm is an ONNX Runtime Web build used for local image analysis.
The manifest declares no permissions and no host permissions. Nothing in the bundle asks for cookie access, history, or broad host access.
What the findings actually show
There are zero malware findings and zero malware-signature hits. The 46 network findings all land inside one file, assets/imageProcessor.worker-BeCTT-Lm.js, on lines 122 through 1997, under the NET-FETCH, NET-XMLHTTPREQUEST and NET-SOCKET_IO titles. A worker named for image processing that contains fetch and XHR calls is the expected shape for code that decodes and inspects media locally.
The endpoint list is where the noise shows. It contains ai.onnx.training and ai.onnx.preview.training, which are ONNX Runtime operator domain strings, not hosts. It contains accountant.cc, activations.cc, alloc.cc, backprop.cc and approximation.cc, which are C++ module fragments lifted out of minified code. It contains a single concatenated sentence beginning "birthpopuptypesapplyimagebeing", which is a minified identifier rather than a domain. 2fwww.in is a URL-encoded fragment. None of these resolve to a service this extension could contact.
The two obfuscation findings, OBFUSCATION-LARGE_WASM_FILE-c2pa_bg.wasm-0 and OBFUSCATION-LARGE_WASM_FILE-assets/ort-wasm-simd-threaded.asyncify-9GUf3Unn.wasm-0, are size heuristics. They fire because WebAssembly binaries are large and not human-readable. A compiled WASM module has no readable source to begin with, so calling it obfuscated says nothing about what it does.
The counterargument
A skeptic would point at the 452 IoC hits and 70 code-smell findings and ask why a two-user extension carries so much machinery, and whether an unknown publisher's package deserves trust. That is a fair question about adoption, not about intent. C2PA parsing and ONNX inference cannot be written in a few kilobytes of hand-tuned JavaScript; they ship as compiled binaries. Bundled dependencies multiply scanner output, because every string in every library is a candidate match. The signals that matter are the empty permission lists, the total absence of malware signatures, and the fact that every network finding sits inside a worker file consistent with the extension's stated job.
Verdict
The evidence describes a legitimate Adobe tool misread by heuristics tuned for minified JavaScript and compiled binaries.
Key Reasons
- Zero malware and zero malware-signature findings across the scan; the highest severity recorded is medium.
- All 46 network findings are confined to assets/imageProcessor.worker-BeCTT-Lm.js, a bundled image worker consistent with the extension's stated function.
- The obfuscation hits are LARGE_WASM_FILE size heuristics on c2pa_bg.wasm and ort-wasm-simd-threaded.asyncify-9GUf3Unn.wasm, both compiled upstream libraries.
- Extension declares no permissions and no host permissions, so it has no standing access to browsing data.
- Listed endpoints such as ai.onnx.training, alloc.cc and the concatenated 'birthpopuptypes...' string are ONNX operator names, C++ fragments and minified identifiers, not resolvable hosts.
False Positive Considerations
- IoC extractor producing 452 domain-shaped matches from minified JS and WASM string tables
- LARGE_WASM_FILE obfuscation heuristic firing on compiled WebAssembly binaries
- Bundled dependency code in assets/ multiplying findings across every library included
- code-smell rules matching generic patterns in non-trivial bundled worker JavaScript
Reviewed 2026-10-15 · version 1.0.2; recommended action: suppress false positive; model confidence 85%.
Chrome version history
Risk trend by version
2 analyzed versions. Each point is the latest successful scan for that version; failed zero-score scans are hidden. Dates: First seen by Risky Plugins.
Pick any point on the chart, then open the Source tab to read that version's code.
Source Code Not Available
Source code is not available for this version of the extension.
About This Extension
Frequently Asked Questions
Similar Extensions
Related extensions from the same publisher or marketplace