Study finds AI browsers remain vulnerable to prompt injection

Researchers found prompt injection attacks still succeed against AI browsers from top vendors despite multiple security guardrails, and say no perfect fix exists.