Spaces:
Sleeping
Sleeping
File size: 13,645 Bytes
788dd2e | 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 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 187 188 189 190 191 192 193 194 195 196 197 198 199 200 201 202 203 204 205 206 207 208 209 210 211 212 213 214 215 216 217 218 219 220 221 222 223 224 225 226 227 228 229 230 231 232 233 234 235 236 237 238 239 240 241 242 243 244 245 246 247 248 249 250 251 252 253 254 255 256 257 258 259 260 261 262 263 264 265 266 267 268 269 270 271 272 273 274 275 276 277 278 279 280 281 282 283 284 285 286 287 288 289 290 291 292 293 294 295 296 297 298 299 300 301 302 303 304 305 306 307 308 309 310 311 312 313 314 315 316 317 318 319 320 321 322 323 324 325 326 327 328 329 330 331 332 333 334 335 336 337 | """
Attack Engine
=============
Reactive adversary that generates attacks based on curriculum level,
observes defender actions, and adapts its strategy.
ImmunoOrg 2.0 - Phase 1: Supports both template-based and LLM-driven adversaries
"""
from __future__ import annotations
import random
from typing import Any
from immunoorg.models import (
Attack, AttackVector, LogEntry, LogSeverity, NetworkNode,
)
from immunoorg.network_graph import NetworkGraph
from immunoorg.llm_adversary import LLMAdversary
# Attack templates by difficulty level
ATTACK_TEMPLATES: dict[int, list[dict[str, Any]]] = {
1: [
{"vector": AttackVector.SQL_INJECTION, "severity": 0.4, "stealth": 0.2,
"description": "Single-point SQL injection on exposed database port"},
{"vector": AttackVector.XSS, "severity": 0.3, "stealth": 0.3,
"description": "XSS on web application"},
{"vector": AttackVector.CREDENTIAL_STUFFING, "severity": 0.35, "stealth": 0.4,
"description": "Credential stuffing on login endpoint"},
],
2: [
{"vector": AttackVector.LATERAL_MOVEMENT, "severity": 0.6, "stealth": 0.5,
"description": "Lateral movement from web tier to app tier"},
{"vector": AttackVector.PRIVILEGE_ESCALATION, "severity": 0.65, "stealth": 0.4,
"description": "Privilege escalation after initial foothold"},
{"vector": AttackVector.PHISHING, "severity": 0.5, "stealth": 0.6,
"description": "Spear phishing targeting management endpoints"},
],
3: [
{"vector": AttackVector.RANSOMWARE, "severity": 0.8, "stealth": 0.3,
"description": "Ransomware deployment with lateral spread"},
{"vector": AttackVector.SUPPLY_CHAIN, "severity": 0.75, "stealth": 0.7,
"description": "Supply chain compromise via dependency injection"},
{"vector": AttackVector.DDOS, "severity": 0.6, "stealth": 0.1,
"description": "DDoS to create distraction for data exfil"},
],
4: [
{"vector": AttackVector.APT_BACKDOOR, "severity": 0.9, "stealth": 0.9,
"description": "APT campaign with persistent backdoor and C2 channels"},
{"vector": AttackVector.ZERO_DAY, "severity": 0.95, "stealth": 0.8,
"description": "Zero-day exploit chain targeting multiple services"},
{"vector": AttackVector.SUPPLY_CHAIN, "severity": 0.85, "stealth": 0.85,
"description": "Multi-stage supply chain attack with delayed activation"},
],
}
class AttackEngine:
"""Generates and manages attacks with reactive adversary behavior.
Supports two modes:
- Template-based (default): Uses fixed attack templates
- LLM-driven: Uses reasoned attack planning with network analysis
"""
def __init__(
self,
network: NetworkGraph,
difficulty: int = 1,
seed: int | None = None,
use_llm_adversary: bool = False,
):
self.network = network
self.difficulty = difficulty
self.rng = random.Random(seed)
self.active_attacks: list[Attack] = []
self.contained_attacks: list[Attack] = []
self.attack_history: list[dict[str, Any]] = []
self.defender_actions_observed: list[str] = []
self.adaptation_counter: int = 0
self.use_llm_adversary = use_llm_adversary
# Initialize LLM adversary if enabled
self.llm_adversary: LLMAdversary | None = None
if use_llm_adversary:
self.llm_adversary = LLMAdversary(network, difficulty, seed)
def generate_initial_attack(self, sim_time: float) -> Attack:
"""Generate the initial attack for an episode."""
if self.use_llm_adversary and self.llm_adversary:
# Use LLM-driven adversary
attack = self.llm_adversary.generate_next_attack(sim_time)
target_node = attack.target_node
self.active_attacks.append(attack)
self.attack_history.append({
"time": sim_time,
"event": "initial_attack",
"vector": attack.vector.value,
"target": target_node,
"description": f"LLM-planned: {attack.metadata.get('rationale', 'N/A')}",
"plan_id": attack.metadata.get("plan_id"),
})
# Compromise the target node
target = self.network.get_node(target_node)
if target:
self.network.compromise_node(target_node, attack.vector, sim_time)
return attack
else:
# Use template-based adversary (original behavior)
templates = ATTACK_TEMPLATES.get(self.difficulty, ATTACK_TEMPLATES[1])
template = self.rng.choice(templates)
# Pick target node based on attack vector
target = self._select_target(template["vector"])
attack = Attack(
vector=template["vector"],
source_node="external",
target_node=target.id if target else "",
entry_point=self._find_entry_point(target, template["vector"]),
severity=template["severity"],
started_at=sim_time,
stealth=template["stealth"],
lateral_path=[target.id] if target else [],
)
# Compromise the target node
if target:
self.network.compromise_node(target.id, template["vector"], sim_time)
self.active_attacks.append(attack)
self.attack_history.append({
"time": sim_time,
"event": "initial_attack",
"vector": template["vector"].value,
"target": target.id if target else "unknown",
"description": template["description"],
})
return attack
def _select_target(self, vector: AttackVector) -> NetworkNode | None:
"""Select an appropriate target node for the attack vector."""
nodes = self.network.get_all_nodes()
if not nodes:
return None
# Vector-specific targeting
preference_map = {
AttackVector.SQL_INJECTION: ["data"],
AttackVector.XSS: ["web"],
AttackVector.CREDENTIAL_STUFFING: ["web", "management"],
AttackVector.LATERAL_MOVEMENT: ["app"],
AttackVector.PRIVILEGE_ESCALATION: ["app", "management"],
AttackVector.PHISHING: ["management"],
AttackVector.RANSOMWARE: ["data", "app"],
AttackVector.DDOS: ["web", "dmz"],
AttackVector.APT_BACKDOOR: ["management", "app"],
AttackVector.SUPPLY_CHAIN: ["app"],
AttackVector.ZERO_DAY: ["dmz", "web"],
}
preferred_tiers = preference_map.get(vector, ["app"])
candidates = [n for n in nodes if n.tier in preferred_tiers and not n.compromised and not n.isolated]
if not candidates:
candidates = [n for n in nodes if not n.compromised and not n.isolated]
if not candidates:
return None
# Prefer nodes with higher vulnerability
candidates.sort(
key=lambda n: max((p.vulnerability_score for p in n.ports), default=0),
reverse=True,
)
# Weighted random from top candidates
top = candidates[:max(1, len(candidates) // 2)]
return self.rng.choice(top)
def _find_entry_point(self, node: NetworkNode | None, vector: AttackVector) -> str:
"""Find the entry point (port/service) for the attack."""
if not node or not node.ports:
return "unknown"
open_ports = [p for p in node.ports if p.status == PortStatus.OPEN]
if open_ports:
most_vulnerable = max(open_ports, key=lambda p: p.vulnerability_score)
return f"{most_vulnerable.service}:{most_vulnerable.port_number}"
return "unknown"
def adversary_tick(self, sim_time: float) -> list[Attack]:
"""Adversary takes a reactive step — propagates attacks, adapts strategy."""
new_attacks: list[Attack] = []
for attack in self.active_attacks:
if attack.contained:
continue
# Propagate laterally based on difficulty
if self.difficulty >= 2:
newly_compromised = self.network.propagate_attack(
attack.target_node, attack, sim_time
)
for nc in newly_compromised:
attack.damage_dealt += 0.1
# Accumulate damage
attack.damage_dealt += attack.severity * 0.02
# At higher difficulties, launch follow-up attacks
if self.difficulty >= 3 and self.rng.random() < 0.05 * self.difficulty:
new_attack = self._launch_followup_attack(sim_time)
if new_attack:
new_attacks.append(new_attack)
# Reactive adaptation based on observed defender actions
if self.adaptation_counter >= 3 and self.difficulty >= 2:
self._adapt_strategy()
self.adaptation_counter = 0
return new_attacks
def observe_defender_action(self, action_name: str) -> None:
"""Adversary observes what the defender does and adapts."""
self.defender_actions_observed.append(action_name)
self.adaptation_counter += 1
# If using LLM adversary, also notify it
if self.llm_adversary:
self.llm_adversary.observe_defender_action(action_name)
def _adapt_strategy(self) -> None:
"""Adapt attack strategy based on observed defender patterns."""
recent = self.defender_actions_observed[-5:]
# If defender is blocking ports, pivot to credential-based attacks
if "block_port" in recent:
for attack in self.active_attacks:
if not attack.contained:
attack.stealth += 0.1
# If defender is isolating nodes, speed up lateral movement
if "isolate_node" in recent:
for attack in self.active_attacks:
if not attack.contained:
attack.severity += 0.05
def _launch_followup_attack(self, sim_time: float) -> Attack | None:
"""Launch a follow-up attack exploiting a different vector."""
used_vectors = {a.vector for a in self.active_attacks}
available = [v for v in AttackVector if v not in used_vectors]
if not available:
return None
vector = self.rng.choice(available)
target = self._select_target(vector)
if not target:
return None
attack = Attack(
vector=vector,
source_node="external",
target_node=target.id,
entry_point=self._find_entry_point(target, vector),
severity=0.3 + self.difficulty * 0.15,
started_at=sim_time,
stealth=0.3 + self.difficulty * 0.1,
lateral_path=[target.id],
)
self.network.compromise_node(target.id, vector, sim_time)
self.active_attacks.append(attack)
self.attack_history.append({
"time": sim_time, "event": "followup_attack",
"vector": vector.value, "target": target.id,
})
return attack
def contain_attack(self, attack_id: str, sim_time: float) -> bool:
"""Mark an attack as contained."""
for attack in self.active_attacks:
if attack.id == attack_id:
attack.contained = True
attack.contained_at = sim_time
self.contained_attacks.append(attack)
return True
return False
def get_active_attacks(self) -> list[Attack]:
return [a for a in self.active_attacks if not a.contained]
def get_total_damage(self) -> float:
return sum(a.damage_dealt for a in self.active_attacks)
def get_adversary_rationale(self) -> str:
"""Get the reasoning behind the adversary's current strategy."""
if self.llm_adversary:
return self.llm_adversary.get_attack_rationale()
return "Template-based adversary (no reasoning available)"
def generate_harder_attack(self, sim_time: float, org_weaknesses: list[str]) -> Attack:
"""Generate a harder attack for the self-improvement loop."""
# Use org weaknesses to pick attack vector
weakness_vector_map = {
"no_devsecops": AttackVector.SUPPLY_CHAIN,
"slow_approval": AttackVector.RANSOMWARE,
"silo_security_engineering": AttackVector.LATERAL_MOVEMENT,
"weak_monitoring": AttackVector.APT_BACKDOOR,
"excessive_trust": AttackVector.PHISHING,
}
vector = AttackVector.APT_BACKDOOR
for weakness in org_weaknesses:
if weakness in weakness_vector_map:
vector = weakness_vector_map[weakness]
break
target = self._select_target(vector)
attack = Attack(
vector=vector,
source_node="external",
target_node=target.id if target else "",
entry_point=self._find_entry_point(target, vector),
severity=min(1.0, 0.5 + self.difficulty * 0.15),
started_at=sim_time,
stealth=min(1.0, 0.4 + self.difficulty * 0.15),
lateral_path=[target.id] if target else [],
)
if target:
self.network.compromise_node(target.id, vector, sim_time)
self.active_attacks.append(attack)
return attack
# Need PortStatus imported
from immunoorg.models import PortStatus
|