Spaces:
Sleeping
Sleeping
File size: 3,086 Bytes
3aeaf3d | 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 61 62 63 64 65 66 67 68 69 70 71 72 73 74 75 76 77 78 79 80 81 82 83 84 85 86 87 88 89 90 91 92 93 94 95 96 97 98 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 | from __future__ import annotations
from dataclasses import dataclass, field
from enum import Enum
from typing import Any
class AttackStrategy(str, Enum):
PERSONA_MANIPULATION = "persona_manipulation"
INDIRECT_INJECTION = "indirect_injection"
MULTI_TURN_ESCALATION = "multi_turn_escalation"
STEERING_VECTOR = "steering_vector"
SOFT_PROMPT_GCG = "soft_prompt_gcg"
REPE_BELIEF_FLIP = "repe_belief_flip"
CIRCUIT_BYPASS = "circuit_bypass"
COALITION_DISTRACTION = "coalition_distraction"
COALITION_POISONING = "coalition_poisoning"
class BlueActionType(str, Enum):
MONITOR = "monitor"
PROBE = "probe"
FLAG = "flag"
BLOCK = "block"
PATCH = "patch"
EXPLAIN = "explain"
@dataclass
class Explanation:
threat_level: str
detection_method: str
layer_implicated: int | None = None
direction_match: str | None = None
evidence_turns: list[int] = field(default_factory=list)
anomaly_score: float = 0.0
recommended_action: str = "warn"
circuit_hypothesis: str | None = None
@dataclass
class RedAction:
strategy: AttackStrategy
sub_strategy: str
payload: str
target_layer: int | None = None
direction_label: str | None = None
magnitude: float = 0.5
coalition_partner: str | None = None
@dataclass
class BlueAction:
action_type: BlueActionType
session_id: str
layer: int | None = None
explanation: Explanation | None = None
patch_reference: str = "clean"
@dataclass
class Turn:
turn_number: int
user_input: str
assistant_output: str
activation_snapshot: dict[int, Any] | None = None
@dataclass
class SessionState:
session_id: str
turns: list[Turn] = field(default_factory=list)
attack_active: bool = False
attack_strategy: AttackStrategy | None = None
attack_payload_turn: int | None = None
extraction_succeeded: bool = False
detected: bool = False
detection_turn: int | None = None
patched: bool = False
@dataclass
class BlueResources:
probe_budget: int = 10
probes_used: int = 0
log_sample_rate: float = 0.5
latency_ticks: int = 1
@dataclass
class TargetSystemState:
hidden_secret: str
policy_rules: list[str]
sessions: dict[str, SessionState]
activation_baseline: dict[int, list[float]]
tick: int = 0
stage: int = 1
seen_strategy_embeddings: list[list[float]] = field(default_factory=list)
blue_resources: BlueResources = field(default_factory=BlueResources)
ground_truth_attacks: dict[str, AttackStrategy] = field(default_factory=dict)
@classmethod
def sample(
cls,
secrets_bank: Any,
rules_bank: Any,
baseline: dict[int, list[float]],
num_sessions: int = 8,
) -> "TargetSystemState":
sessions = {f"sess_{i}": SessionState(session_id=f"sess_{i}") for i in range(num_sessions)}
return cls(
hidden_secret=secrets_bank.sample(),
policy_rules=rules_bank.sample(n=5),
sessions=sessions,
activation_baseline=baseline,
)
|