AI Transparency Companies Are Not Being Honest With Each Other Let Alone the Public

Industry's Supposed Commitment to Openness Masks Widespread Deception and Evasion\n\nCompanies that position themselves as pro-transparency in the AI industry "can't even be honest with each other, let alone the rest of us," according to a scathing analysis of AI industry claims about openness and accountability.\n\n### The Problem\n\n- AI companies claiming transparency are deceiving each other and the public\n- Supposed openness commitments often theatrical rather than substantive\n- Companies hide critical information about training data, model capabilities, and safety testing\n- Industry self-regulation proving inadequate\n\n### Key Hypocrisies\n\n- Training data: Companies refuse to disclose what data their models are trained on, citing competitive concerns\n- Safety testing: Internal safety evaluations remain proprietary, preventing independent verification\n- Capabilities: Companies selectively report benchmark results to present models in the best light\n- Incident reporting: Safety incidents are often downplayed or not disclosed at all\n\n### The Transparency Theater\n\nMany AI companies have signed voluntary commitments to transparency and safety. However, the gap between these public commitments and actual practice is enormous. The report suggests that much of what passes for transparency is actually carefully managed PR.\n\n### What's Needed\n\nThe analysis argues for mandatory transparency requirements, independent audits, and standardized disclosure frameworks that would make it impossible for companies to selectively hide information.\n\nSource: The Verge, Jess Weatherbed