Websites Are Planting Hidden Unicode Traps for AI Bots

Sites and forums have started embedding hidden prompt-injection payloads and invisible Unicode steganography to detect and break AI agents during scraping and signup. Bruce Schneier said an AI agent emailed him about these traps.

• Sites hide prompt injections and invisible Unicode to catch bots
• Humans never see the traps, but LLMs parsing the DOM fall for them
• An agent emailed Bruce Schneier about the traps it found
• The defense works only while agents blindly follow page text