What happened
On October 29, SAG-AFTRA, the American Federation of Musicians (AFM), and both branches of the Writers Guild of America (WGA) filed a motion for leave to submit a 34-page friend-of-the-court brief in *In re OpenAI, Inc., Copyright Infringement Litigation*. The case is before Judge Sidney H. Stein in the Southern District of New York. The proposed brief supports the newspaper and author plaintiffs and asks the court to reject a rule treating AI training as fair use across the board. The unions are not plaintiffs, and the court must grant their request before the brief formally joins the record.
The filing’s labor argument is not limited to writers and publishers. It says harms from unlicensed training can reach professionals whose work is used in commercial media, and points to voice actors as a concrete example. The brief describes performers’ work sustaining characters and other vocal performances through long recording sessions, and connects synthetic voices to lost bookings. It cites reports of actors being hired for app-based vertical dramas—short episodic series distributed through phone apps—only to learn that AI had replaced them.
The unions cite the National Association of Voice Actors’ State of Voiceover in 2026 survey. The survey ran from January 22 to February 22 and collected 1,379 responses, distributed through social media, voiceover groups on Facebook, and NAVA’s contact list. Asked whether they had knowingly lost a job to a synthetic voice, 21% answered yes. The source also reports that 13% had willingly agreed to let a third party make a synthetic version of their voice, while 53% said they would be open to having one ethically created in the future. The brief says 94% had not made a synthetic version of their voice and cites industry impact, limited AI laws, and consent among respondents’ reasons.
The filing argues that licensing markets already exist and points to union bargaining. It says agreements that followed the 2023 strikes established minimum compensation terms for digital replicas and guardrails around fully synthetic characters. According to the brief, 2026 contracts also provide notice and bargaining rights for third-party licenses of covered work used to train public-facing, commercially available generative AI systems. OpenAI and Microsoft argue that model training falls within fair use; the unions argue the question should not receive a categorical answer. The source gives no price or licensing rate.
What the archive adds
This is a new development in a story the archive has covered. Fair use benefits everyone described OpenAI’s argument for fair use and support it cited in the New York Times case. The union motion adds a labor-side brief opposing a blanket rule, with voice actors’ reported job losses as evidence. Other related filings include House Judiciary Democrats’ brief and Wikimedia’s brief; they make distinct arguments in the broader litigation.
Trend signal
This filing reinforces the broader pattern of AI disputes being argued through competing claims about who bears the costs and who captures value. But no supplied SITE_PATTERNS thesis specifically covers labor representation or fair-use litigation, so this development does not map cleanly to a pattern page.
Why now
The motion was filed October 29 while the consolidated copyright case is weighing whether training on protected works can qualify as fair use. The unions say their members’ livelihoods depend on the same kinds of rights at issue, and that current contract provisions show bargaining over training licenses is already happening. The court has not yet decided whether to accept their brief.
Connections to other events
Earlier coverage in the archive includes an OpenAI fair-use argument and separate amicus filings by House Judiciary Democrats and Wikimedia. The unions’ motion joins that broader set of interventions but advances a different emphasis: labor, performers’ work, and negotiated terms for digital replicas and training licenses. The supplied source also ties the filing to the 2023 strikes and 2026 contracts, without establishing how the court will treat those agreements.
Why it matters
For teams building models, this filing is a reminder that the training-data question is not only about whether a legal doctrine can permit use without permission. Labor groups are asking the court to consider market effects on workers and point to negotiated licensing terms as an existing alternative. If courts take such arguments seriously, builders may face more pressure to identify the rights and people implicated by training sources, document licensing decisions, and plan for negotiation rather than assume a categorical fair-use position.
The evidence here is not a compliance rule or a court ruling. Still, it makes voice and performance data a tangible example of the stakes. Teams using synthetic voices or training on recorded material should distinguish what the filing asserts from what has been established, while tracking the case and the contracts that may define practical expectations.
Limits and uncertainty
The unions’ brief is a proposed filing, and the source says the court must grant leave before it becomes part of the record. Its arguments are advocacy, not findings by Judge Stein. The NAVA survey records respondents’ answers; its distribution through social media, Facebook groups, and NAVA’s contact list does not establish that the sample represents all voice actors. The 21% figure is self-reported and does not quantify total jobs lost, identify every cause of a lost booking, or show how often synthetic voices were trained on any particular work. The source does not report a decision on the motion or the case’s merits.
Reader takeaway
The union filing makes the OpenAI training dispute more explicitly about labor as well as publishers and authors. Its strongest concrete evidence is a survey in which 21% of 1,379 voice-actor respondents said they had knowingly lost work to a synthetic voice. That is a reported survey result and an argument before the court—not a judicial finding or a population-wide estimate.