Chatbots argue against election conspiracy theories, then willingly illustrate them

Details

The Brennan Center for Justice published research testing whether major AI platforms enforce consistent safety guardrails across modalities. The findings reveal a clear gap: text-based chatbots from major AI providers generally refused to validate false election claims, but the image generation tools from those same providers readily produced realistic-looking fabricated evidence for identical misinformation. This modality inconsistency is the core policy finding. Companies appear to have invested more heavily in guardrails for conversational AI than for image generation, even though fabricated images are often more persuasive and shareable than text-based claims. The study's implications are pointed: an AI company can have responsible-seeming chatbots while its image tools are generating political disinformation on demand. This creates a false impression of platform safety that regulators and users relying on chatbot behavior alone would miss. The source article is a brief summary; the Brennan Center's full report likely contains specific companies tested, image types generated, and methodological details.