The proposal treats undisclosed output steering as presumptively deceptive. If an AI company prioritizes ideological, commercial, or compliance-driven objectives over truthful responses without telling users, it violates the FTC Act. Companies can avoid liability only through clear and conspicuous disclosure that their AI prioritizes certain objectives over user requests. The FTC rests on a straightforward premise: consumers reasonably expect AI to produce truthful, accurate outputs and have no reason to assume hidden ideological manipulation.
The action traces to a December 2025 executive order directing the FTC to address state laws requiring AI model alteration. The proposal specifically targets Colorado's Artificial Intelligence Act as impliedly preempted where it forces companies to alter outputs for state ideological objectives. The framework invokes three principles from the AI Life Cycle Core Principles: Truth, Transparency, and Accountability.
Attorneys should monitor this closely. The proposal establishes a new consumer-protection framework that treats undisclosed model alignment—including techniques like reinforcement learning from human feedback—as a legal liability regardless of motive. This shifts enforcement risk substantially. As the comment period remains open, industry stakeholders are analyzing how the disclosure requirement will reshape AI vendor risk and development strategy. The outcome will likely influence how companies document and justify model tuning decisions.