Prompt injection in HTTP headers: what 217M web responses show
A study examined 217 million HTTP responses to identify prompt‑injection attempts targeting language models. Researchers found that malicious actors embed crafted prompts in headers such as User‑Agent, Referer and Cookie, causing downstream AI services to follow unintended instructions. The analysis highlights the scale of the issue and suggests stricter header sanitization to protect LLM‑powered applications for future.