Publishers Weekly: Authors Guild updates AI best practices for writers

The Authors Guild updated its AI best practices on May 12, 2026, releasing revised recommendations to help writers understand the legal, contractual, and ethical dimensions of AI use in their work. The update was reported by Publishers Weekly and adds new sections that were not present in earlier guidance.

The revised document has four sections: Guiding Principles; Risks to Be Aware of When Using AI; Categories of Use; and Recommended Best Practices. The Guild frames the guidance explicitly as a spectrum rather than a list of prohibitions. The underlying position is that not all AI use raises the same concerns, and writers need a way to distinguish cases rather than a blanket rule that applies equally to, say, using AI to generate metadata and using AI to write a manuscript chapter.

Two legal risks receive particular attention. First, AI-generated text currently lacks copyright protection. A manuscript that includes undisclosed AI-generated content may create problems in copyright applications, and submitting such content without disclosure could constitute fraud. Second, the Guild notes that major language models were trained on published works without authors’ permission, compensation, or the ability to opt out — a structural concern that sits behind how individual writers choose to use these tools today.

Contract implications are also addressed directly. Many publishing agreements include warranties that submitted work is the author’s own. Authors who include AI-generated material without disclosure may be in breach of those warranties, depending on specific contract language. The guidance names this risk rather than leaving it implicit.

The document does not prohibit AI use. It provides a framework for identifying which uses carry legal risk, which carry ethical complexity, and which are relatively uncontroversial. For writers navigating contracts, copyright applications, and editorial relationships that were established before the current generation of AI tools existed, the updated document gives specific language for understanding where caution is warranted.