A short blog post comparing Aaron Swartz’s prosecution with Meta’s acquisition of books for AI training reached 934 points and 209 comments on Hacker News by early August 21. That unusually fast response matters more than the post’s most incendiary wording. It shows how readily developers recognize a power gap: one programmer faced federal felony charges after downloading academic papers, while one of the world’s largest companies has kept shipping AI systems after torrenting far more data.
The comparison is emotionally direct and legally imprecise. The post argues that Swartz was prosecuted for essentially the same conduct Meta has carried out with little consequence. But Swartz was charged under computer-access and fraud laws, not copyright law. Meta is fighting civil copyright claims about copying books for Llama and allegedly redistributing pieces of them through BitTorrent. Those are different acts, statutes and procedural stages.
Correcting the analogy does not make the underlying concern disappear. It makes the disparity easier to locate: in prosecutorial discretion, the difference between criminal and civil enforcement, and the years a well-funded company can spend contesting what an individual had to confront personally.
What the Swartz prosecution was actually about
Swartz, who helped develop RSS 1.0 and later became a prominent open-access activist, used MIT’s network to download a large part of JSTOR’s archive in 2010 and 2011. JSTOR’s account says 4.8 million articles were downloaded, roughly 80 percent of its database at the time. MIT’s later review says the activity happened in several waves and bypassed attempts to stop it. A laptop was eventually found connected to the network in a basement wiring closet.
The government did not indict Swartz for infringing copyright. The Justice Department’s 2011 announcement listed wire fraud, computer fraud, unlawfully obtaining information from a protected computer, and recklessly damaging a protected computer. Prosecutors alleged that he evaded technical blocks, impaired JSTOR’s systems and intended to distribute a substantial part of the archive. Those allegations were never tested at trial. Swartz died by suicide in January 2013 while the case was pending.
The same DOJ announcement said the charged counts carried an aggregate maximum of 35 years in prison and a $1 million fine. That was a statutory ceiling, not a forecast sentence, but it still illustrates the pressure built into a federal felony case. By then the stakes were not an abstract copyright dispute between publishers and a downloader. They included arrest, criminal defense, possible incarceration and forfeiture.
JSTOR had already recovered the downloaded material and secured assurances that it would not be distributed. In its statement on the case, the archive said prosecution was the government’s decision and that, after recovering the material, it had no interest in an ongoing legal matter. That did not end the federal case.
Calling this simply a prosecution for “scraping” therefore omits the government’s central theory: unauthorized access accompanied by repeated evasion and alleged system damage. Whether that theory justified the breadth and intensity of the prosecution is a separate question. Many critics have long argued that it did not. Accuracy does not require accepting the prosecutors’ choices.
Meta’s torrenting sits in a different legal lane
The facts established in Kadrey v. Meta are serious, but not identical. In a June 2025 order, US District Judge Vince Chhabria wrote that there was no dispute Meta torrented datasets from LibGen and Anna’s Archive. Meta then added downloaded books to datasets used to train Llama models.
BitTorrent complicates the case because downloading and uploading can happen together. The order says a Meta engineer wrote a script intended to prevent seeding, while the parties dispute whether Meta still reuploaded data through the protocol, how much it uploaded and whether any pieces came from the named plaintiffs’ books. That distinction is central. Copying a work into a training dataset and distributing copies to other peers implicate separate exclusive rights under copyright law.
Meta won summary judgment in 2025 on the 13 named authors’ claim that copying their books for model training infringed copyright. Yet the ruling was deliberately narrow. Chhabria wrote that it did not establish that Meta’s training use was generally lawful; those plaintiffs had failed to build evidence around the market-harm argument the court considered most important. The result applied to those authors on that record, not to every book or every AI training pipeline.
The litigation is also not over. A March 2026 order describes three distinct theories: copying for training, direct infringement through alleged BitTorrent uploads, and contributory infringement based on facilitating redistribution by other peers. The court allowed the authors to add the third claim. It had not yet decided the upload-related claims on summary judgment.
Meta itself confirms the remaining exposure. In its June 2026 quarterly filing, the company says copyright and contributory-infringement claims concerning alleged distribution during downloading remain active. Summary-judgment arguments are scheduled for February 25, 2027. The filing also lists several newer cases making similar claims.
So “without consequence” overstates what is known. Meta faces discovery, legal expense and potentially significant liability. But “with no comparable personal consequence” is closer to the observable difference. No Meta employee is identified in these sources as facing a Swartz-style federal criminal prosecution over the book downloads. The company continues to develop and distribute Llama while the civil system works through claims years after the conduct.
Scraping is not one legal category
The viral comparison also bundles together several meanings of scraping. Fetching public pages, accessing an authenticated database, bypassing network controls, copying protected expression and redistributing files are not interchangeable. A crawler can implicate contract, privacy, database, trespass or copyright rules without necessarily committing a computer crime.
US computer-access law has narrowed since the Swartz era. In 2021, the Supreme Court’s Van Buren v. United States opinion rejected a broad reading of the Computer Fraud and Abuse Act’s “exceeds authorized access” language. In 2022, the Ninth Circuit held in the hiQ v. LinkedIn dispute that scraping publicly accessible profiles was unlikely to count as accessing computers “without authorization” under that law, though other claims could still apply.
Neither decision retroactively resolves the Swartz case, and neither gives AI companies a universal right to collect whatever they can reach. Meta’s disputed conduct in Kadrey was not merely a crawler reading public web pages. It involved obtaining large book collections from shadow libraries through a peer-to-peer protocol. The decisive questions concern copyright, fair use, proof of distribution and market harm.
For developers building crawlers or training pipelines, the practical lesson is to separate acquisition from use. Record where each dataset came from, what access controls were present, which licence or exception supports copying, and whether the transfer method also sends data to third parties. A model’s output may be transformative while the way its training data was obtained or redistributed creates a different claim. “Publicly reachable,” “permitted to download,” “licensed for training” and “safe to upload” are four different assertions.
The power critique survives the fact-check
The post’s popularity is not evidence that Meta broke a criminal law. Hacker News points measure attention, not adjudication. Still, the response exposes an issue that a narrow legal comparison can miss. Enforcement happens through institutions with choices: prosecutors decide whether to bring charges; rights holders decide whether to sue; judges decide claims based on the evidence the parties manage to present. Money changes a defendant’s ability to endure each stage.
Swartz encountered the state as an individual. Meta encounters authors and publishers as a corporation with specialist counsel, years of discovery, multiple defenses and a product line that can continue operating during the dispute. The difference does not prove unlawful favoritism. It does explain why “the cases are different” is an incomplete answer to the public anger. A system can apply distinct laws correctly and still impose radically different practical burdens.
The next concrete test is not whether a viral analogy wins an argument. It is what the court does with evidence about BitTorrent uploading, whether the remaining claims reach a class or trial, and how the February 2027 summary-judgment hearing treats Meta’s fair-use defense. Until then, the grounded conclusion is narrower than either side’s slogan: Swartz and Meta did not face the same legal case, but the distance between criminal pressure on an individual and slow civil accountability for a corporation remains worth examining.