Don’t trust AI companies with your content. Verify them instead

Recent copyright lawsuits against AI company Anthropic reveal a troubling trend: big tech firms are increasingly accused of stealing content to train their AI models. Sony Music and Warner are suing Anthropic, along with Universal, alleging the company illicitly pirated millions of music tracks and song lyrics. Anthropic has faced multiple allegations, with claims that Anthropic co-founder Benjamin Mann personally directed torrenting activities and openly discussed them in Slack channels.

The company responded by stating they would vigorously defend themselves in court. This lawsuit is a prime example of why tech companies cannot be trusted with content. AI companies often take a liberal view of "fair use" when harvesting content, leading to an all-or-nothing approach. Publishers are often faced with the dilemma of blocking AI bots from accessing their content, which reduces visibility in AI answers, or allowing bots, which may result in their content being used for training or other purposes.

Verification of AI companies' claims is crucial in this situation. Publishers can block training bots while selectively allowing retrieval bots, which are used for real-time search and answers. By using a content distribution network (CDN) to verify bots and monitoring retrieval bots using "tracer" phrases, publishers can ensure their content is not being used for training.

However, this approach requires a measured, piece-by-piece testing strategy to avoid potential pitfalls. Despite the media industry's reasons to be cautious, it's important to note that AI companies have a vested interest in playing by the rules due to the significant financial stakes involved in licensing deals and courtroom settlements.