Aaron Swartz faced 35-year sentence for 70 GB JSTOR scrape; Meta trained Llama on 80 TB torrented books without criminal charges

Swartz used a script to download academic papers from JSTOR under an MIT guest account in September 2010. Prosecutors charged him under the Computer Fraud and Abuse Act and wire fraud statutes, seeking asset forfeiture and up to 35 years imprisonment. He died by suicide in January 2013 before trial. Court records show the government rejected multiple plea offers that included no jail time.

Meta engineers documented the use of Books3 and LibGen torrents containing roughly 80 TB of text for Llama pretraining in internal technical reports released in 2023. No criminal referral occurred. The company faces ongoing civil suits from authors but continues model deployment. Data volume exceeds Swartz downloads by three orders of magnitude with no parallel felony prosecution.

The disparity tracks enforcement patterns: CFAA cases against individuals average 18-month sentences when data access exceeds authorized limits, while corporate ingestion of copyrighted corpora triggers only DMCA takedown notices or licensing negotiations. Meta's scale and revenue model place the activity inside accepted industry practice for foundation model training.

Civil litigation outcomes will set precedent for disgorgement of model weights rather than imprisonment. Expect further complaints from rightsholders citing the same torrent sources, with damages capped at statutory maximums per work rather than felony exposure.