Search papers, labs, and topics across Lattice.
This paper investigates the security implications of using Handlebars templating in large language model prompts, specifically focusing on the structural role injection vulnerability introduced by the choice between double-brace and triple-brace expressions. Through a model-free analysis and extensive trials across various delimiter families and models, the authors reveal that while HTML escaping mitigates some risks, it fails to protect against certain delimiter schemes, allowing for significant attack vectors. The findings indicate that the current escaping mechanisms are insufficient and highlight the need for a more robust separation of instruction and data to enhance security in LLM applications.
HTML escaping in Handlebars prompts offers inadequate protection, leaving critical attack vectors open for structural role injection.
Large language model applications build prompts from templates, and Handlebars is a widely used templating engine and the default prompt-template format in Microsoft Semantic Kernel. Its double-brace {x} expression HTML-escapes the interpolated value and is documented as the safe default; its triple-brace {x} expression inserts the value raw. We show that this choice silently governs an application's exposure to structural role injection, where attacker-controlled data carries chat role delimiters that forge a higher-privilege turn. A model-free analysis establishes the mechanism: Handlebars escaping rewrites angle brackets but not square brackets, colons, or Markdown hashes, so it neutralises ChatML, Llama-3, and XML role delimiters (survival rate 0.00) while leaving Llama-2 [INST], legacy Human:/Assistant:, and Markdown ### delimiters intact (survival rate 1.00 for the last two). We then run 5760 trials across seven delimiter families, two attack objectives, and four models (GPT-3.5 Turbo, GPT-4o mini, GPT-4.1 mini, Claude Haiku 4.5) at a combined API cost of 1.63 USD. GPT-3.5 Turbo follows the task-hijack instruction in 97% of raw and 91% of escaped trials, with the escaping protection concentrated in the angle-bracket families and absent for the colon- and Markdown-based families; the harder secret-exfiltration objective, which does not saturate, exposes the same family interaction more cleanly. Claude Haiku 4.5 resists both objectives almost entirely. The escaped default protects only the delimiter schemes whose characters HTML escaping happens to cover, gives no protection for the rest, and cannot substitute for a structural separation of instruction and data.
