What changed
In March, Hachette Book Group canceled the US publication of Shy Girl after allegations that the novel was largely AI-generated. Pangram, an AI detection tool, claimed 78 percent of the text was AI-generated. Mia Ballard, the author, denied these claims, stating an editor made AI-assisted changes to a previously self-published version. Since then, other authors have faced similar accusations, though they have also denied them.
What this means in practice
Authors now face the risk of being accused of using AI without clear mechanisms to defend themselves. Andrea Bartz, a lead plaintiff in a class-action suit against Anthropic, warns that accusations can lead to a "witch hunt." The Authors Guild, represented by Mary Rasenberger, acknowledges the existence of tools like Workings, Blindproof, Wintertext, Human Ink, and ObraVera, which aim to track and document an author’s writing process. These tools can encrypt and fingerprint written work, or use typing rhythms to create a record of authorship. However, Rasenberger notes that these tools are not foolproof and may produce false positives.
Limitations and availability
The tools available are not perfect. Max Spero, cofounder and CEO of Pangram, admits that "paper trail" tools cannot capture all aspects of a writer’s process, such as hand-typing AI text or using auto-typers. Additionally, the AI arms race means vulnerabilities and fixes are ongoing. ObraVera’s Brady Ridgway explains that his platform can trace reuse of material but cannot confirm whether AI was used in the creation of the text. The Authors Guild has also reported instances where human-authored work was incorrectly flagged as AI-generated. While some argue that vetting systems for human authorship are impractical, Jane Friedman, a publishing expert, suggests that AI scandals may be short-lived due to improvements in detection tools. However, she notes that publishers are unlikely to adopt verification methods that cost extra time or money.
