Spaces:
Sleeping
Sleeping
File size: 10,310 Bytes
0d4249e | 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 | """
EBSysMLSec-Py: LLM-Assisted HAZOP Threat Analyzer
Gradio demo for Hugging Face Spaces.
Demonstrates the AI-assisted HAZOP step from:
Poorhadi & Troubitsyna (IMBSA 2022, RSSRail 2023, SAFECOMP 2024)
Users describe any safety-critical system and receive a structured
HAZOP threat analysis β the formal safety-security interaction analysis
that feeds into Event-B modelling.
"""
import json
import os
from pathlib import Path
import gradio as gr
import pandas as pd
from hazop.hazop_analyzer import analyze_system, INSULIN_PUMP_MODEL
# ββ Pre-loaded threats (cached from the insulin pump analysis) ββββββββββββββ
PRELOADED_THREATS_PATH = Path("hazop/threats.json")
def _load_preloaded() -> list[dict]:
if PRELOADED_THREATS_PATH.exists():
return json.loads(PRELOADED_THREATS_PATH.read_text(encoding="utf-8"))
return []
def _threats_to_df(threats: list[dict]) -> pd.DataFrame:
if not threats:
return pd.DataFrame()
rows = []
for t in threats:
rows.append({
"ID": t.get("id", ""),
"Component": t.get("component", ""),
"Flow": t.get("flow", ""),
"Guide Word": t.get("guide_word", ""),
"Deviation": t.get("deviation", ""),
"Consequence": t.get("consequence", ""),
"Attack Vector": t.get("attack_vector", ""),
"Violated Invariant": t.get("violated_invariant", "none"),
"Severity": t.get("severity", ""),
"Event-B Event": t.get("event_b_attack_event", "none"),
"Attack Machine": t.get("attack_machine", "none"),
})
return pd.DataFrame(rows)
def _severity_summary(threats: list[dict]) -> str:
counts = {"critical": 0, "high": 0, "medium": 0, "low": 0}
for t in threats:
sev = t.get("severity", "low")
counts[sev] = counts.get(sev, 0) + 1
parts = [f"**{k.upper()}**: {v}" for k, v in counts.items() if v > 0]
return f"Total threats: **{len(threats)}** β " + " | ".join(parts)
# ββ Default system description (shown on load) βββββββββββββββββββββββββββββββ
DEFAULT_SYSTEM = """\
System: Autonomous Insulin Pump Controller
Components:
- GlucoseSensor: reads blood glucose level; states: IDLE, MEASURING, TRANSMITTING, ERROR
- DoseCalculator [AttackSurface]: computes insulin dose; states: WAITING, COMPUTING, DONE, ERROR
- SafetyMonitor: validates dose against safety bounds; states: MONITORING, CHECKING, APPROVED, REJECTED
- PumpActuator: delivers insulin; states: IDLE, PRIMING, DELIVERING, DONE
- NetworkInterface [AttackSurface]: external communication; states: IDLE, RECEIVING, TRANSMITTING
- PatientProfile: stores patient parameters (MAX_SAFE_DOSE=50, HYPO_THRESHOLD=70, MIN_BATTERY=10)
Information Flows:
- F1: GlucoseSensor β DoseCalculator (GlucoseReading)
- F2: PatientProfile β DoseCalculator (PatientParams)
- F3: DoseCalculator β SafetyMonitor (DoseRequest)
- F4: SafetyMonitor β PumpActuator (DoseCommand)
- F5: NetworkInterface β DoseCalculator (ExternalCmd) [ATTACK SURFACE FLOW]
Safety Invariants (Event-B):
- INV1: delivered_dose β€ MAX_SAFE_DOSE
- INV2: delivered_dose > 0 β glucose_reading β₯ HYPO_THRESHOLD
- INV3: delivered_dose > 0 β battery_level β₯ MIN_BATTERY_LEVEL
- INV4: delivered_dose > 0 β command_approved = TRUE
- INV5: dose_request β€ MAX_SAFE_DOSE\
"""
def _parse_freetext_model(text: str) -> dict:
"""
Convert the freetext system description into the model_info dict
that hazop_analyzer.analyze_system() expects.
"""
lines = [l.strip() for l in text.strip().splitlines() if l.strip()]
name = "System"
components, flows, invariants = [], [], []
section = None
for line in lines:
if line.startswith("System:"):
name = line.replace("System:", "").strip()
elif line.lower().startswith("component"):
section = "components"
elif line.lower().startswith("information flow") or line.lower().startswith("flow"):
section = "flows"
elif line.lower().startswith("safety invariant"):
section = "invariants"
elif line.startswith("-"):
item = line[1:].strip()
if section == "components":
is_attack = "[AttackSurface]" in item or "[attacksurface]" in item.lower()
comp_name = item.split(":")[0].replace("[AttackSurface]", "").strip()
desc = item.split(":", 1)[1].strip() if ":" in item else ""
components.append({"name": comp_name, "is_attack_surface": is_attack, "description": desc})
elif section == "flows":
# Try to parse "F1: A β B (Signal)" or "A β B via Signal"
parts = item.split(":", 1)
flow_id = parts[0].strip() if len(parts) > 1 else f"F{len(flows)+1}"
rest = parts[1].strip() if len(parts) > 1 else item
is_attack = "[ATTACK" in rest.upper()
rest_clean = rest.split("[")[0].strip()
if "β" in rest_clean:
src_dst, *sig_parts = rest_clean.split("(")
src, dst = src_dst.split("β")
signal = sig_parts[0].rstrip(")").strip() if sig_parts else "Data"
flows.append({
"id": flow_id,
"source": src.strip(),
"target": dst.strip(),
"signal": signal,
"flow_type": "attack" if is_attack else "normal",
})
elif section == "invariants":
if ":" in item:
inv_name, inv_text = item.split(":", 1)
invariants.append({"name": inv_name.strip(), "text": inv_text.strip()})
return {"name": name, "components": components, "flows": flows, "invariants": invariants}
# ββ Gradio callbacks ββββββββββββββββββββββββββββββββββββββββββββββββββββββββββ
def run_hazop_analysis(system_text: str, use_preloaded: bool):
"""Main callback: run HAZOP and return (DataFrame, summary, JSON)."""
if use_preloaded:
threats = _load_preloaded()
source_note = "Pre-generated threats from the insulin pump analysis (no API call made)."
else:
api_key = os.environ.get("ANTHROPIC_API_KEY", "")
if not api_key:
return (
pd.DataFrame(),
"β οΈ Set the ANTHROPIC_API_KEY environment variable to run live analysis.",
"",
)
model_info = _parse_freetext_model(system_text)
try:
threats = analyze_system(model_info)
except Exception as e:
return pd.DataFrame(), f"Error: {e}", ""
source_note = f"Live LLM analysis ({len(threats)} threats generated)."
df = _threats_to_df(threats)
summary = _severity_summary(threats) + f"\n\n*{source_note}*"
raw_json = json.dumps(threats, indent=2, ensure_ascii=False)
return df, summary, raw_json
# ββ Layout ββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββ
with gr.Blocks(
title="EBSysMLSec-Py β LLM HAZOP Analyzer",
theme=gr.themes.Soft(),
) as demo:
gr.Markdown("""
# EBSysMLSec-Py: LLM-Assisted HAZOP Threat Analyzer
Reproduces and extends the safety-security interaction analysis methodology of
**Poorhadi & Troubitsyna** (IMBSA 2022 Β· RSSRail 2023 Β· SAFECOMP 2024).
**What this does:** Applies the seven HAZOP guide words (NO, MORE, LESS, AS WELL AS, PART OF, REVERSE, OTHER THAN)
to each information flow in your SysML model, producing structured threat scenarios that feed directly
into Event-B formal verification (the `ATK_*` events in the `.bum` files).
**Full repo:** `sysml/` β `translator/` β `eventb/` β `verification/` β see the GitHub repository.
""")
with gr.Row():
with gr.Column(scale=1):
system_input = gr.Textbox(
label="System Description",
info="Describe your system: components, information flows, safety invariants",
value=DEFAULT_SYSTEM,
lines=22,
max_lines=40,
)
use_preloaded = gr.Checkbox(
label="Use pre-generated results (no API key needed)",
value=True,
)
analyze_btn = gr.Button("Run HAZOP Analysis", variant="primary", size="lg")
with gr.Column(scale=2):
summary_md = gr.Markdown("*Run the analysis to see results.*")
threats_df = gr.Dataframe(
label="HAZOP Threat Table",
wrap=True,
interactive=False,
)
with gr.Accordion("Raw JSON output", open=False):
json_out = gr.Code(language="json", label="threats.json")
gr.Markdown("""
---
### How results connect to the formal model
| HAZOP threat | Event-B artifact |
|---|---|
| `T-001` (MORE on F1) | `ATK_SpoofGlucoseReading` in `Attack_Spoofing.bum` β INV2 fails |
| `T-009` (OTHER THAN on F4) | `ATK_InjectDeliveryCommand` in `Attack_Injection.bum` β INV4 fails |
| `T-013` (OTHER THAN on F5) | `ATK_ReplayHighDoseCommand` in `Attack_Replay.bum` β INV1 fails |
Each failing proof obligation in Rodin corresponds to a row in this table where **Violated Invariant β none**.
### Reference
Poorhadi, E., Troubitsyna, E., DΓ‘n, G. (2024). *Automating an Integrated Model-Driven Approach to Analysing the Impact of Cyberattacks on Safety.* SAFECOMP 2024. DOI: 10.1007/978-3-031-68738-9_5
""")
analyze_btn.click(
fn=run_hazop_analysis,
inputs=[system_input, use_preloaded],
outputs=[threats_df, summary_md, json_out],
)
demo.load(
fn=lambda: run_hazop_analysis(DEFAULT_SYSTEM, True),
outputs=[threats_df, summary_md, json_out],
)
if __name__ == "__main__":
demo.launch()
|