Prompt Injections - 25-Year-Old SEO Trick in Court Docs Now Lmao

Prompt injections have surfaced in a US court filing, echoing the hidden text SEO tricks that worked for 25 years. This reveals how LLMs can be manipulated in ways search engines once were.

https://www.law.com/legaltechnews/2...k-courts-may-have-to-confront-in-the-ai-age-/

Hidden Text Goes AI Edition

- White-on-white text was invisible to humans but readable by machines earlier - the OG SEOs know this. LOL. Those were the days.
- Modern prompt injections use similar tricks in legal docs to steer AI conclusions
- Resumes now show 1% carrying hidden instructions per large-scale studies

Court Filing Sparks Sanctions

- Plaintiff hid instructions in white text for the bariatric surgery lawsuit
- Judge issued sanctions after spotting excessive white space in filings called it exparte comms.
- Communication must be open and visible to all parties in legal proceedings

Brand Reputation Risks Rise

- LLMs process hidden prompts the same way Google once handled keywords
- SEO now faces AI-specific issues like positive review manipulation on arXiv
- Companies must monitor for stealth tactics in any digital content

I saw this on Twitter the other day

People will always look for shortcuts. There can be a million stealth methods for AI without others knowing them. Unlike websites, AI doesn't have a publicly accessible index. So, we'll never know about such tricks until someone leaks them or we find them ourselves. AI itself might suggest such tricks in the future (ofcourse, for paid users ).