Infosecurity Europe: Prompt Injection Remains Unsolved, OWASP Researcher Warns

Ariel Fogel, speaking at Infosecurity Europe 2026, warns that prompt injection is an unresolved architectural problem for agentic AI: because LLMs process inputs as a single token sequence, privilege boundaries between system prompts, user input and retrieved content are hard to enforce. The report highlights the 'Lethal Trifecta' (access to private data, untrusted content, and external communication) and Meta’s 'Rule of Two' as heuristics, and urges defenders to adopt machine-speed monitoring, automated containment, tighter identity/session design, and cross-disciplinary incident playbooks to reduce risk.