The Rare Book Scandal: How a Crypto AI Project Is Destroying Digital Artifacts for Training Data

Reality check: Over the past 30 days, the on-chain footprint of a single AI training pipeline has consumed 1,200+ rare NFT books. The metadata is gone. The original tokens are burnt. What remains is a vectorized ghost in a large language model. Let’s look at the numbers. The project in question—call it BookChain AI—claims to be building the world’s largest treasury of high-quality book text for AI training. According to my manual audit of 285 transactions on Ethereum, the project has systematically purchased rare, verified NFT editions of out-of-print literary works, transferred them to a burn address, and then extracted the raw text from the token metadata. The original NFTs are destroyed. The data is now private. Context: BookChain AI is a DAO with a token called $BOOK that has a market cap of $42 million. The project’s whitepaper, which I read in full, describes a “data pipeline” that acquires rare digital books through NFT marketplaces, decrypts the embedded text, and feeds it into a proprietary LLM training set. The justification: “to preserve and democratize access to rare literature.” But the on-chain evidence tells a different story. The burn rate of NFTs is 100%. The text is not made public. The only access is through a future API, priced in $BOOK. Core: I traced the flow of 1,200 NFTs from acquisition to destruction. The pattern is clinical. The project uses a multisig wallet (0xdead…beef) to buy NFTs from collections like “RareBookDAO” and “LiteraryGenesis.” Within 24 hours, the NFTs are sent to a burn contract. The text metadata—stored in the token URI as IPFS hashes—is then fetched by a private server. The IPFS pins are removed. The original artworks (illustrations, cover designs) are discarded. Only the raw text survives. Two things stand out. First, the average price paid per NFT is $8.50, far below their last sale price. This suggests the project is using a bot to snipe undervalued, illiquid assets. Second, the burn contract emits a log event that includes a cryptographic hash of the extracted text. That hash is not indexed anywhere. It’s a dead end for verification. Numbers don’t lie, but they can be buried. I ran a simple statistical test: the distribution of NFT purchase prices follows a Poisson-like pattern, peaking at $5–$10. This is consistent with automated buying, not organic collection. The project is not preserving culture; it’s strip-mining it. Contrarian: Some argue that destroying the original NFT is necessary to prevent copyright disputes—if the token is burned, there is no longer a public copy that could be claimed as infringing. That logic is flawed. Code is law. Bugs are fatal. Burning the token does not erase the chain of provenance. The burn transaction is visible forever. The original metadata hashes are still on IPFS, even if unpinned. And the text itself is now embedded in a model that may regurgitate it. The project is creating a different kind of liability: training data memorization. Moreover, the project’s tokenomics rely on scarcity. The $BOOK token is used to access the API. By destroying the original NFTs, the project artificially creates a monopoly on that text. This is not democratization. It’s rent-seeking. Hype dies. Math survives. The math says they are burning assets at a loss to create a data moat that may never be valuable enough to recover the cost. Takeaway: The next signal to watch is the project’s next funding round. If they raise a Series A at a $500 million valuation, we will know that institutional investors are betting on data scarcity. But if the on-chain NFT burn rate continues without a corresponding increase in API usage, the token price will collapse. Follow the gas, not the news. The real story is not the destruction of rare books—it’s the unsustainable economics of extracting data from a finite supply of digital artifacts. The question is not whether the project is ethical. The question is whether the numbers will ever add up. For deep analysis, I will continue to track the burn wallet and cross-reference it with known LLM training outputs. If you hold $BOOK, consider the risk of regulatory action. The FTC is watching. The author societies are watching. And I am watching.