CAA is demanding Meta overhaul its privacy protections for Muse Image, a new AI image generator that creates synthetic photos using public Instagram handles. The talent agency's rebuke centers on Meta's opt-out model, which requires users to affirmatively block the platform from harvesting their data rather than securing consent upfront.

Muse Image lets anyone generate AI-created photos by inputting someone's Instagram username. The tool trains on publicly available data, but CAA argues the burden should fall on Meta to get permission before using images for AI training, not on individuals to prevent it after the fact. This inverts standard privacy frameworks that entertainment and talent industries have fought to establish.

The pushback reflects broader anxieties about generative AI's relationship to creative work and performer likenesses. Studios, unions, and agencies fear uncontrolled AI systems could scrape talent's images without compensation or approval, undermining residuals and contractual protections actors fought for during recent strikes. The WGA and SAG-AFTRA negotiations included specific guardrails around synthetic performers and AI-generated content using member likenesses.

Meta's opt-out approach treats privacy as something users must defend themselves, rather than something the platform defaults to protecting. For celebrities and public figures with large follower bases, opting out individually becomes cumbersome. For lesser-known performers, the process remains invisible unless they discover it independently.

CAA's statement amplifies pressure on Meta as regulators and creative communities scrutinize how tech companies deploy generative AI. The Federal Trade Commission has signaled interest in enforcement actions against companies that misuse biometric data or violate privacy commitments. Meta faces ongoing reputation challenges around user data practices following years of privacy scandals.

The Muse Image dust-up signals that Hollywood's AI concerns extend beyond traditional production workflows. Talent representation now contends with unauthorized synthetic imagery of their