{"id":"3a3e2480-8068-43ec-8218-d1f3c54a0fc6","arxiv_id":"2507.06926","paper_version":1,"verdict":"CONDITIONAL","confidence":"MODERATE","novelty_score":4.0,"correctness_risk":"medium","formal_verification":"none","parameter_count":1,"one_line_summary":"An empirical review of 16,000+ ERC-721 NFT metadata records finds sparse copyright information, storage fragility, and unreliable artist identification, leading to a negative verdict on NFTs for Australian copyright protection.","lead":"This paper examines NFT metadata from the Ethereum blockchain to test whether NFTs can protect Australian and Indigenous artists' copyright. It finds very few records with copyright terms, fragile storage links, and unreliable artist names, so the authors conclude NFTs are not ready for this purpose.","discovery_kind":"new_application","skeptic_critique":{"model":"deepseek-v4-flash","headline":"The artist-identification pillar is undermined by a substring matcher that cannot handle name order or pseudonyms; the 'very hard' result may reflect algorithm failure rather than missing metadata.","rationale":"The reader's weakest assumption correctly identifies the dirty name list as a problem, but the more load-bearing defect is the matching algorithm's false-negative bias. A dirty list primarily inflates the raw match count via false positives; correcting it would lower the count, which actually supports the 'very hard' conclusion. The false-negative bias, by contrast, could mask a large number of true Australian artists simply because their names do not appear contiguously in register form. Since the paper explicitly presents artist identification as a key capability for the Australian Copyright Agency (Section 1, RQ3), this methodological gap directly weakens one of the three pillars of the central claim. The paper deserves credit for collecting a substantial new dataset (43,810 ERC-721 tokens, 16,784 acquired metadata) and for the quantitative evidence on copyright/license scarcity (only ~203 NFTs with such information), which is largely independent of the identification method. Those findings support a conditional negative verdict on current NFT readiness. However, the internal numeric inconsistencies and the identification flaw mean the paper cannot be accepted as-is; the claimed counts and the RQ3 answer must be re-derived with a sound matching procedure and a cleaned register. Since the reader already issued a CONDITIONAL verdict and our concern reinforces that conditionality rather than overturning the overall qualitative conclusion, the verdict remains unchanged.","tokens_in":9687,"tokens_out":11919,"duration_ms":134368,"concrete_test":"Re-run artist identification on the same 13,462 metadata strings using a cleaned version of the Copyright Agency register (retaining only entries that are plausibly human names, with at least a first and last token) and a token-based matcher that checks whether any first-name+last-name pair from the register appears in the metadata, allowing for name-order variation (e.g., 'Serwah Attafuah' matches 'Attafuah, Serwah Ama Gyekyewah Bianca'). Also search dedicated creator/artist fields if present. Manually inspect a random sample of 100 matches to compute precision, and report recall against a hand-checked subset of NFTs known to be by Australian artists.","verdict_should_be":"UNCHANGED","load_bearing_attack":"The RQ3 conclusion rests on a crude identification procedure: each artist's full name (or name without middle name) from the Australian Copyright Agency register is searched as a contiguous substring in the NFT metadata string. This produces both false positives and false negatives. False positives are visible in Table 2: 'What' (103), 'ME' (10), 'Pcd' (4), 'Ipg' (3), and 'Mama' (3) are not artist names, yet they account for the majority of the 199 reported matches. The authors acknowledge this in Section 5.3, but they treat the dirty list as one of the reasons identification is hard rather than as a flaw in their own benchmark. More importantly, the method creates systematic false negatives. As the paper itself notes, the register lists 'Attafuah, Serwah Ama Gyekyewah Bianca' while the NFT metadata uses 'Serwah Attafuah'; a substring search on full names will never match these because the token order and intervening names break the substring. The paper attributes this to the artist using a different name, but it is actually a failure of the matching algorithm. Consequently, the answer to RQ3 ('it is very hard to identify artists') conflates the limitations of a naive substring matcher with the absence of identifying information in NFT metadata. The central claim's third pillar is therefore not established. Additional internal inconsistencies (Table 2 counts sum to 198, not 199; percentages sum to >100%; Section 5.1 uses 16,769 acquired metadata while Section 4 reports 16,784) reinforce that the quantitative results cannot be taken at face value.","agreement_with_reader":"partial"},"referee_report":{"model":"deepseek-v4-flash","summary":"This paper empirically studies ERC-721 NFT metadata and artwork to assess whether NFTs are ready to protect Australian and Indigenous artists' copyright. The authors collect contract addresses from Ethereum blocks 16,000,000 to 21,269,338, identify 43,810 ERC-721 contracts, retrieve and decode metadata from decentralized, centralized, and on-chain storage, and analyze copyright/license keywords, storage distributions, duplication, artwork accessibility, and Australian artist identification by substring matching against the Australian Copyright Agency's name list. They report that only 1.21% of metadata contains copyright or license information, that storage and pointer failures create security risks, and that artist identification is very hard, concluding that NFTs are not ready for this purpose.","tokens_in":10008,"tokens_out":5739,"duration_ms":53241,"significance":"The paper addresses a timely and policy-relevant question, and its empirical approach—using a self-built Erigon node, collecting real on-chain data, and separating metadata and artwork storage types—is a strength that could inform standards discussions. The concrete quantitative findings, such as the low prevalence of license terms and the concentration of storage on IPFS, are potentially useful if the data are cleaned and reported consistently. However, the manuscript as written contains several load-bearing data inconsistencies and a flawed artist-identification benchmark, so the strength of the conclusion is not yet supported.","major_comments":[{"comment":"The RQ3 conclusion that 'it is very hard to identify artists from NFT metadata' is not established by the reported experiment. The top matches in Table 2 come from entries in the Australian Copyright Agency list that are not artist names ('What', 'ME', 'Pcd', 'Ipg', 'Mama'), accounting for 123 of the 199 reported matches, and the substring matcher cannot handle name order or intervening names, as the paper's own example of 'Attafuah, Serwah Ama Gyekyewah Bianca' versus 'Serwah Attafuah' demonstrates. Section 5.3 mentions these issues but treats them as properties of the data rather than as artifacts of the matching procedure and the dirty ground-truth list, so the answer to RQ3 conflates matcher failure with missing metadata.","section":"§5.3, Table 2"},{"comment":"The duplication statistics do not add up. The text reports 1,159 duplicated metadata, but the bar values in Figure 4 sum to 746, and 14,466 non-duplicated plus 1,159 duplicated equals 15,625, not the stated 16,784 acquired metadata. The authors should reconcile these numbers or correct them.","section":"§4, Figure 4"},{"comment":"The denominator used for the central 1.21% claim is inconsistent across the paper. Section 4 reports 16,784 acquired metadata and 13,462 correct metadata, while Section 5.1 uses 16,769; the artist identification is performed on 13,462. The authors need to define 'correct metadata' and state exactly which denominator supports each percentage.","section":"§4 and §5.1"},{"comment":"The 'Common Terms(%)' column in Table 1 sums to 115%, not 100%, which makes the reported license-type distribution unreliable. The authors should recompute or relabel these values.","section":"Table 1"},{"comment":"The copyright/license filtering step is under-specified. The text says that after searching for 'copyright' and 'license', some metadata 'has one keyword but does not have any copyright or license information' and is removed, but no criterion is given for making that determination. Since the set of 203 NFTs is the basis for the paper's main quantitative conclusion, this step needs a precise, reproducible definition.","section":"§3 and §5.1"}],"minor_comments":[{"comment":"There are several typos, including 'empericially' and 'unfortnately' in the abstract and 'inclcues' in Section 3; these should be corrected.","section":"Abstract, §3"},{"comment":"Table 2 is internally inconsistent: the 'Total numbers' column sums to 198, not 199; the percentage for 'Bandi' is 17.4757% but 18/199 is 9.05%; and the percentages in the table sum to more than 100%.","section":"Table 2"},{"comment":"The paper alternately refers to 43,810 ERC721 'tokens' and 43,810 contract addresses; the authors should clarify whether the unit is contracts or tokens.","section":"§3, §4"},{"comment":"The security discussion cites numbers such as 570 wrong-domain cases in decentralized storage and 574 wrong URLs, but these are not clearly traceable to Figure 8 or any table; the authors should make the mapping between text and figure explicit.","section":"§5.2, Figure 8"},{"comment":"The phrase 'correct metadata' is used to report 13,462 items, but the paper earlier defines only 'acquired metadata'; a precise definition of 'correct' is needed.","section":"§4"}],"recommendation":"major_revision","confidential_remarks":"The empirical dataset and the research question are suitable for a security- or cryptoeconomics-oriented venue, but the manuscript needs a major methodological rewrite before it can be accepted. I would encourage the editor to require a corrected dataset table and a re-analysis of RQ3 with a validated name-matching approach, rather than treating the current reported numbers as reliable."},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"The paper is a mixed bag. The central claim—that current ERC-721 NFTs are not ready to serve as a copyright protection tool for Australian and Indigenous artists—is likely true and consistent with what we already know about NFT metadata fragility. The new dataset and the Australian case study are genuinely useful. But the quantitative evidence is not reliable enough to carry the argument.\n\nThe strongest part is the empirical collection itself: 43,810 ERC-721 contracts identified from a local Erigon node, with metadata and artwork storage classified into on-chain, IPFS/Arweave, and centralized services. The finding that only about 1.2% of metadata contains copyright or license terms is the most valuable takeaway. The security discussion, while not novel, is well grounded in the storage breakdown.\n\nUnfortunately, there are serious internal inconsistencies. The paper reports 13,462 \"correct metadata\" in Section 4, then 16,784, then 16,769 in Section 5.1. Duplication counts in Figure 4 sum to 746 tokens, but the text claims 1,159 duplicated metadata. These aren't minor typos; they undermine the denominator for every percentage in the paper.\n\nThe artist identification analysis is the weakest pillar. The method is a naive substring search of names from the Australian Copyright Agency register against metadata strings. The register itself contains non-names like \"What\" and \"ME\", which account for 113 of the 199 reported matches. The authors note this but still treat the raw results as evidence that identification is \"very hard.\" Worse, the matching algorithm creates systematic false negatives: a full name like \"Attafuah, Serwah Ama Gyekyewah Bianca\" will never match \"Serwah Attafuah\" in the metadata. So RQ3's conclusion conflates algorithm failure with absence of identifying information. That's a real problem.\n\nThe citation pattern is fine. Self-citations appear for background, but the empirical results don't depend on prior theories. The informal tone in places (\"empirically\" misspelled in the abstract, duplicated sentences in Section 5.1) suggests the paper was rushed.\n\nWho is this for? Policy researchers and practitioners in the Australian art world who want a quick look at NFT metadata. It could also be a cautionary example for empirical blockchain studies. But with the current numerical inconsistencies and the flawed matching, I would not cite the numbers as reliable. A serious referee should ask for corrected counts, a cleaned name list, and either a more sophisticated matching method (e.g., token-based fuzzy matching) or an explicit statement that the identification result is only a lower bound.\n\nRecommendation: send it to peer review, but expect the reviewers to require substantial revision before any of its quantitative claims are usable.","headline":"New Australian NFT dataset marred by arithmetic errors and a naive name-matching step that undermines the artist-identification conclusion.","tokens_in":10495,"tokens_out":3067,"would_cite":false,"duration_ms":31260,"reading_group":"maybe","serious_thinker":"yes","would_accept_peer_review":true},"rs_alignment":null,"lean_confirmation":null,"pith_extraction":{"msc":[],"pacs":[],"model":"deepseek-v4-flash","headline":"An empirical audit of ERC-721 NFTs concludes that current token metadata cannot reliably protect Australian artists' copyrights, with only 203 of 16,769 metadata records containing any copyright or license information.","keywords":["NFT","ERC-721","copyright","metadata","artist identification","Australian artists","Indigenous artists","blockchain security"],"falsifier":"A definitive check would be to re-run the identification pipeline on the same 13,462 metadata records with a cleaned artist register that excludes non-name tokens and includes known pseudonyms; if the corrected match rate is high, the paper's 'very hard to identify' conclusion fails. A second check is to sample newly minted ERC-721 tokens after the introduction of a standardized license field: if a large majority of fresh records carry license data, the 1.21% figure is a snapshot of an early template, not a property of NFTs.","tokens_in":9524,"feed_emoji":"🎨","tokens_out":4896,"duration_ms":43517,"temperature":0.7,"pith_summary":"The paper tests whether NFTs, as currently built on the ERC-721 standard, could serve as a copyright-protection and artist-identification tool for Australian artists, including Indigenous artists. It collected 16,784 successfully retrieved metadata records from Ethereum mainnet NFTs and analyzed copyright, license, security, and artist-name fields. It finds that only about 1.21% of those records contain any copyright or license information, and that artist identification from metadata is unreliable, both because metadata lacks geographic or standardized creator fields and because the author-name list used as ground truth contains errors. The intended conclusion is that the NFT ecosystem is not yet ready to safeguard Australian artists' copyrights or to let the Australian Copyright Agency reliably identify artists on-chain.","feed_headline":"Only 1.21% of NFT metadata carries copyright or license data","feed_subtitle":"An audit of 16,784 Ethereum ERC-721 tokens shows why Australian artists cannot yet rely on NFTs to prove ownership or be identified.","key_machinery":"The central object is the ERC-721 NFT metadata record: a JSON document that a token's on-chain tokenURI points to, commonly following OpenSea's template with name, description, image, and attributes fields. The paper's method extracts these records from 43,810 ERC-721 contracts on Ethereum mainnet, decodes on-chain encodings (Base64, UTF-8, ASCII), fetches off-chain metadata from IPFS, Arweave, and cloud platforms, then scans the JSON strings for the keywords 'copyright' and 'license' and for artist names drawn from the Australian Copyright Agency list. This keyword-and-substring scan on a large real-world sample is the mechanism that produces the 1.21% copyright figure and the 199-artist identification list, so the whole argument stands or falls on whether this metadata, and the name list, faithfully represent what artists put into NFTs.","core_discovery":"The paper claims that NFTs are not ready to protect Australian and Indigenous artists' copyright. The evidence is empirical: of 16,769 acquired metadata records from ERC-721 tokens, only 203 (1.21%) contain copyright or license information, and those records scatter the information across four different metadata locations, with no standard placement. A keyword search for artist names from the Australian Copyright Agency's list found 199 matching tokens in 13,462 correct metadata records, but substring matching produced false positives such as 'What' and 'ME', and the authors themselves note that the name list is incomplete and contains incorrect entries. The paper further reports security issues: missing metadata pointers, inaccessible artwork URLs, and tampering risks in centralized and decentralized storage, while on-chain storage resists tampering but is rarely used. The authors conclude that current NFTs have significant limitations for copyright protection and that identifying Australian artists from metadata is 'very hard.'","pith_inferences":["This reading extends the paper: a clean test of its conclusion would be to run the same identification pipeline against a curated, deduplicated artist registry rather than the Australian Copyright Agency list; if identification rates rise sharply, the bottleneck is the registry, not NFT metadata.","The paper's low copyright rate may partly reflect OpenSea's early metadata template, which had no dedicated copyright field; platform-driven template changes could quickly shift the statistics without changing the underlying blockchain.","The findings suggest a cost-geometry argument: storing full copyright declarations on-chain is expensive, so off-chain standardized metadata with on-chain hashes is the cheapest path to the protection the paper wants.","For Indigenous artists, the absence of protocols for communal or cultural ownership in ERC-721 metadata is a deeper gap than the paper's counting exercise shows."],"forward_implications":["If NFTs are to protect Australian artists, metadata standards must require machine-readable copyright and license fields in a fixed location.","Artists and platforms need to include geographic or nationality data and canonical creator names in metadata to make identification feasible.","On-chain storage or hashed pointers are needed to prevent metadata and artwork tampering.","Copyright agencies cannot currently rely on NFT metadata for artist identification; doing so would misattribute works.","Future standards should separate copyright ownership from license terms and follow structured formats like OpenSea attributes."],"supporting_citations":[{"why":"Supplies the working definition of NFTs and the taxonomy of on-chain/off-chain storage that the empirical collection uses.","marker":"[13]"},{"why":"Argues that metadata tampering and invalid caching break NFT-to-provenance links, motivating the security analysis.","marker":"[7]"},{"why":"Documents off-chain NFT hijacking via centralized components, the security baseline the paper's storage-type comparison extends.","marker":"[12]"},{"why":"Shows fragility of the NFT-to-asset link due to poor decentralized storage and duplication, which the paper re-observes in its pointer failures.","marker":"[15]"},{"why":"Proposes image hash functions for detecting duplicated NFTs, the detection approach the paper's duplication counts complement.","marker":"[6]"},{"why":"Maps marketplace design flaws and scams in the NFT ecosystem, used as background for the security discussion.","marker":"[3]"}],"fun_headline_variants":["NFTs not ready for Australian artists, audit finds","Only 1.21% of NFTs include copyright or license data","NFT copyright gaps fail Australian artists, study says","Artists beware: NFTs don't protect Aussie copyright","Indigenous artists exposed: NFT copyright still shaky"],"cache_read_input_tokens":3200,"weakest_assumption_plain":"The paper's artist-identification result depends on treating the Australian Copyright Agency's name list, which contains non-name entries like 'What' and 'ME', as the ground-truth register of Australian artists.","fun_headline_variants_meta":{"raw":{"variants":["NFTs not ready for Australian artists, audit finds","Only 1.21% of NFTs include copyright or license data","NFT copyright gaps fail Australian artists, study says","Artists beware: NFTs don't protect Aussie copyright","Indigenous artists exposed: NFT copyright still shaky"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.000254,"raw_usage":{"total_tokens":1536,"prompt_tokens":884,"completion_tokens":652,"prompt_tokens_details":{"cached_tokens":384},"prompt_cache_hit_tokens":384,"prompt_cache_miss_tokens":500,"completion_tokens_details":{"reasoning_tokens":573}},"tokens_in":500,"tokens_out":652,"duration_ms":7074,"temperature":1.0,"reasoning_tokens":573,"cache_read_input_tokens":384,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-06T18:51:03.070314+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"A definitive check would be to re-run the identification pipeline on the same 13,462 metadata records with a cleaned artist register that excludes non-name tokens and includes known pseudonyms; if the corrected match rate is high, the paper's 'very hard to identify' conclusion fails. A second check is to sample newly minted ERC-721 tokens after the introduction of a standardized license field: if a large majority of fresh records carry license data, the 1.21% figure is a snapshot of an early template, not a property of NFTs.","supporting_citations":[{"cited_title":"In: International Conference on Financial Cryp- tography and Data Security (FC) (2023)","cited_arxiv_id":null,"evidence_quote":"Documents off-chain NFT hijacking via centralized components, the security baseline the paper's storage-type comparison extends."},{"cited_title":"In: Proceedings of the ACM Web Conference (WWW) (2023)","cited_arxiv_id":null,"evidence_quote":"Shows fragility of the NFT-to-asset link due to poor decentralized storage and duplication, which the paper re-observes in its pointer failures."},{"cited_title":"In: IEEE International Conference on Blockchain and Cryptocurrency (ICBC) (2024)","cited_arxiv_id":null,"evidence_quote":"Proposes image hash functions for detecting duplicated NFTs, the detection approach the paper's duplication counts complement."},{"cited_title":"In: ACM SIGSAC Conference on Computer and Commu- nications Security (CCS) (2022)","cited_arxiv_id":null,"evidence_quote":"Maps marketplace design flaws and scams in the NFT ecosystem, used as background for the security discussion."}],"review_version":1}