ASCII smuggling shifts from AI prompt attacks to spam filter evasion

Techub reported that ASCII smuggling, a technique previously used to hide malicious prompts in attacks on AI agents, is now being used by spammers to slip past email platform filters designed to catch mass spam campaigns. The method drew attention two years ago because it made prompt injection attacks harder to spot. In practice, malicious instructions are embedded in emails or other untrusted content meant for large language models through a series of special Unicode tags. Those encoded characters can be read by computers, yet remain almost entirely invisible to human readers. That split allows a language model to pick up the hidden instructions while a person reading the same message may not notice anything unusual. Techub attributed the report to Ars Technica.