Amazon Rufus Guardrail Jailbreak via ASCII Decimal Encoding
A guardrail jailbreak vulnerability was discovered in Amazon Rufus and other LLM implementations using ASCII decimal encoding to bypass inference restrictions. The technique allows attackers to circumvent safety controls designed to prevent provision of information hazards and law-breaking content.
What happened
A guardrail jailbreak vulnerability was discovered affecting Amazon Rufus and other LLM implementations. The vulnerability uses ASCII decimal encoding combined with strategic obfuscation to bypass inference restrictions.
Reported impact
The jailbreak allows attackers to bypass guardrails designed to prevent providing information hazards and instructions for law-breaking activities. The disclosure categorizes the risk as Level 3: Significant Risk, with potential health impacts and significant regulatory interest.
Index analysis
The attack employs encoding techniques, specifically ASCII decimal encoding, to obscure malicious inputs from the model's detection mechanisms. The technique falls under 'Other Encoding' strategies that use unconventional encoding schemes to evade content filters.
Classification
Sources and evidence
This record was researched and written by the Index, based on the following original reporting.