Anthropic has told investors it cannot guarantee that its efforts to respect website access controls always work, and says it may not be able to verify whether training material was copyright-protected. The disclosure puts a practical question behind the IPO noise: what could the company owe if its legal defence fails?
Anthropic Watch analysis
What happened
The Telegraph reports that Anthropic’s stock-market filing says its efforts to observe paywalls and robots.txt, which tells web crawlers what pages they can access, “may not always be successful”. The filing also says the company might be unable to verify whether material used to train its models was protected by copyright.
Anthropic reportedly warned investors that if it could not rely on fair use, it might have to pay “significant fees” to obtain licences. The paper says the filing also disclosed two Canadian publishers’ lawsuits alleging their journalism was used to train Claude. Those are allegations, not findings that Anthropic infringed copyright. Read The Telegraph’s report. The article says Anthropic was contacted for comment; it does not report a response.
Why it matters
The issue is not only whether a particular use of material is lawful. If licensing becomes necessary, the cost of training and operating models could change, while publishers and creators have a direct stake in how their work is used. The filing’s uncertainty is a disclosure of risk, not a ruling on copyright law or proof that any particular work was used unlawfully.
Our read
The important news is the company’s reported admission of uncertainty, not the publishers’ moral verdict. For investors, creators and AI users, the question is whether courts, licensing agreements or future business practices turn that uncertainty into a measurable cost. Copyright law is not usually a thrilling read, but the bill attached to it might be.
What to watch
- Whether Anthropic provides further detail about how it checks training material and honours access controls.
- What the Canadian cases allege and how Anthropic responds.
- Whether licensing or court decisions change the cost of acquiring training material.
Discussion spark: If an AI company cannot reliably confirm whether training material is protected by copyright, should it have to license that material before using it, or should the law leave the question to courts?
Sources and evidence
- Anthropic lacks ‘basic moral compass’, say publishers – The Telegraph (20 November 2026, 14:43 UTC)
Anthropic Watch is independently operated by WittyWires. It is not affiliated with, endorsed by, or operated by Anthropic.