File size: 1,606 Bytes
eaae571 | 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 | {
"taxonomy_version": "1.0",
"project": "prompt_injection_security_eval",
"attack_surfaces": [
{
"attack_surface": "direct",
"description": "User-supplied prompts that try to override or redirect model instructions.",
"families": [
"direct_instruction_override",
"goal_hijacking",
"roleplay_bypass"
]
},
{
"attack_surface": "indirect",
"description": "Injected instructions hidden inside external content such as retrieved documents or tool output.",
"families": [
"retrieved_content_injection",
"context_smuggling",
"semantic_separation"
]
},
{
"attack_surface": "leakage",
"description": "Attempts to expose hidden prompts, policies, internal instructions, or private context.",
"families": [
"system_prompt_extraction",
"policy_disclosure",
"instruction_leakage"
]
},
{
"attack_surface": "adaptive",
"description": "Multi-turn or iterative attacks that change based on model responses.",
"families": [
"iterative_refinement",
"response_adaptation",
"multi_turn_probe"
]
},
{
"attack_surface": "tool_mediated",
"description": "Attacks that target tool-using or agentic workflows.",
"families": [
"tool_abuse",
"action_hijacking",
"environment_manipulation"
]
}
],
"labels": [
"prompt_leakage",
"instruction_override",
"goal_hijacking",
"unsafe_tool_action",
"benign",
"failed_attack"
]
}
|