What are AI transparency requirements?

At a glance

• Disclose the bot: tell customers a chatbot or voice assistant is AI, not a human, unless obvious^[5].
• Label AI content: mark AI-generated or altered images, audio, video, and deepfakes as artificial, often machine-readable^[1].
• Reveal the inputs: California’s AB 2013 makes public generative-AI developers publish a training-data summary^[3].
• 2026 deadlines are live and apply to anyone serving those markets, wherever you are based.

What you must disclose

Three buckets: tell customers when they’re talking to AI^[5], mark anything your AI generates or alters^[1], and (for generative-AI makers) publish training-data details^[3]. The EU AI Act’s Article 50 covers the first two; California’s AB 2013 drives the third.

Who and by when

Rules split between “providers” who build the AI and “deployers” who use it on customers; a small shop with an off-the-shelf chatbot is usually a deployer. Deadlines: California AB 2013 (Jan 1, 2026)^[3], Colorado AI Act (Feb 1)^[4], EU Article 50 (Aug 2)^[2].

Why it matters

Penalties are real and active: EU fines reach EUR 35M or 7% of global turnover^[2], US states treat violations as deceptive trade practices, and the FTC is already suing firms over hidden AI claims^[6]. The fix is cheap: add a clear “you’re chatting with AI” notice and label AI-made media.

Bottom line

Never let anyone mistake your AI for a human or your synthetic content for real; adding disclosures now beats a fine later.