White-on-white text that once stuffed keywords into search results is resurfacing as prompt injection: invisible instructions embedded in documents to bias large language models. Recent incidents and studies show models often follow those instructions because they can’t separate control text from content.
Hidden instructions—white text set to match the background—have reappeared across papers, resumes, buttons, and court filings to influence LLM output.
Prompt injections are present in real-world systems at measurable scale: a resume dataset found roughly 1% contained hidden prompts, and legal filings have already used the tactic with sanctions following.
# What happened A technique search marketers used 25 years ago—hiding text on a page so machines see it but people do not—has migrated into the era of large language models. Instead of hiding keywords for ranking, attackers now embed machine-only instructions that tell an LLM how to evaluate, summarize, or act on a document.
# Where hidden prompts have been found
# Why models follow hidden instructions
# What the experiments show A University of Turin team tested many payload types and positions across ChatGPT and Gemini. They ran 100 real papers through five payload families and repeated runs to produce 42,000 outputs. Results:
# Practical implications
# What this means for readers
# Bottom line
A CISO who sees a low CVE count and deprioritizes prompt injection is reading the scoreboard wrong. Prompt injection has held the No. 1 spot on the OWASP Top 10 for LLM Applications for three consecutive years . When two leaders of that list checked it against 6,639 labeled real-world incidents, it came back at No. 12.
This seems to work : Researchers from Tracebit on Monday said they found that placing prompt injections alongside passwords, cryptographic keys, and other secrets stored on Amazon Web Services was often all that was needed to shut down attacks from AI hacking agents. The prompts direct the attacking LLM to perform an a
For the third year in a row, prompt injection tops the OWASP GenAI / LLM Top Ten list issued today as being the most vulnerable practice to be exploited. In previous years, the list was built on the judgment of voters. But, according to Steve Wilson, co-chair of the OWASP GenAI Security Project and Chief... … continue
Prompt injection remains the most dangerous security threat to LLMs, according to OWASP’s latest Top 10 LLM Applications list
AI search only feels like a new sport to the people who were never really training. Here's what actually changed, and what didn't. The post AI Search Only Feels New If Your SEO Was Shallow appeared first on Search Engine Journal .
SEO National™, a search engine optimization firm founded in 2007, is highlighting industry data that reveals the continued dominance of organic search over AI-driven referral traffic. According to Ahrefs’ “AI vs Search Traffic Analysis,” which tracked 101,872 websites through June 2026, traditional search en
Loading more related stories...
Open the app view to save this story, compare related coverage, and continue from the same source.