ChatGPT Doesn't Trust Chargers Fans: Guardrail Sensitivity in Context
This study addresses biases in LLM guardrails, which are critical for fairness in AI applications, though it is incremental as it builds on known issues of model biases.
The paper investigates how contextual user information, such as demographics and sports fandom, biases the refusal guardrails of GPT-3.5, finding that younger, female, and Asian-American personas are more likely to trigger refusals for censored or illegal requests, and guardrails adjust sensitivity based on inferred political ideologies.
While the biases of language models in production are extensively documented, the biases of their guardrails have been neglected. This paper studies how contextual information about the user influences the likelihood of an LLM to refuse to execute a request. By generating user biographies that offer ideological and demographic information, we find a number of biases in guardrail sensitivity on GPT-3.5. Younger, female, and Asian-American personas are more likely to trigger a refusal guardrail when requesting censored or illegal information. Guardrails are also sycophantic, refusing to comply with requests for a political position the user is likely to disagree with. We find that certain identity groups and seemingly innocuous information, e.g., sports fandom, can elicit changes in guardrail sensitivity similar to direct statements of political ideology. For each demographic category and even for American football team fandom, we find that ChatGPT appears to infer a likely political ideology and modify guardrail behavior accordingly.