New research finds OpenAI's safety features for teenagers fail to mitigate critical risks. Despite guardrails launched in August, the chatbot remains vulnerable to bypassing protections. This gap exposes a disconnect between corporate safety claims and actual model behavior.