fix: Switch notebook to OpenAI gpt-4o (#341)

* fix: Switch notebook to OpenAI gpt-4o and read keys from env vars

* fix: Add organization param and fix VaultGuard transforms reference

* fix: Remove hardcoded API keys from notebook and use placeholders

* fix: Add trailing newline to notebook

Author: rdheekonda

examples/airt/agentic_red_teaming_attacks.ipynb

@@ -4,25 +4,7 @@
    "cell_type": "markdown",
    "id": "0",
    "metadata": {},
-   "source": [
-    "# Agentic AI Red Teaming\n",
-    "\n",
-    "Automated adversarial attacks against agentic AI challenges on\n",
-    "[Dreadnode Crucible](https://platform.dreadnode.io) using the AIRT framework.\n",
-    "\n",
-    "| Challenge | Category | Difficulty |\n",
-    "|-----------|----------|------------|\n",
-    "| **toolshed** | DevOps Tool Misuse | Medium |\n",
-    "| **webwhisper** | Indirect Prompt Injection | Medium |\n",
-    "| **vaultguard** | Multi-Agent Defense Bypass | Hard |\n",
-    "\n",
-    "**Attacks**: TAP (beam search), GOAT (graph exploration), Crescendo (progressive escalation)\n",
-    "\n",
-    "```bash\n",
-    "export CRUCIBLE_API_KEY=\"your-api-key\"  # from https://platform.dreadnode.io/account\n",
-    "export GROQ_API_KEY=\"your-groq-api-key\"\n",
-    "```"
-   ]
+   "source": "# Agentic AI Red Teaming\n\nAutomated adversarial attacks against agentic AI challenges on\n[Dreadnode Crucible](https://platform.dreadnode.io) using the AIRT framework.\n\n| Challenge | Category | Difficulty |\n|-----------|----------|------------|\n| **toolshed** | DevOps Tool Misuse | Medium |\n| **webwhisper** | Indirect Prompt Injection | Medium |\n| **vaultguard** | Multi-Agent Defense Bypass | Hard |\n\n**Attacks**: TAP (beam search), GOAT (graph exploration), Crescendo (progressive escalation)"
   },
   {
    "cell_type": "code",
@@ -51,9 +33,10 @@
    "metadata": {},
    "outputs": [],
    "source": [
-    "CRUCIBLE_API_KEY = os.environ[\"CRUCIBLE_API_KEY\"]  # https://platform.dreadnode.io/account\n",
+    "CRUCIBLE_API_KEY = \"your-crucible-api-key\"  # https://platform.dreadnode.io/account\n",
+    "OPENAI_API_KEY = \"your-openai-api-key\"\n",
     "CRUCIBLE_URL = \"https://platform.dreadnode.io\"\n",
-    "ATTACKER_MODEL = \"groq/meta-llama/llama-4-scout-17b-16e-instruct\"\n",
+    "ATTACKER_MODEL = \"openai/gpt-4o\"\n",
     "EVALUATOR_MODEL = ATTACKER_MODEL\n",
     "\n",
     "TOOLSHED_URL = \"https://toolshed.platform.dreadnode.io\"\n",
@@ -63,7 +46,7 @@
     "MAX_TRIALS = 30\n",
     "VAULTGUARD_MAX_TRIALS = 50\n",
     "\n",
-    "dn.configure(server=CRUCIBLE_URL, token=CRUCIBLE_API_KEY, organization=\"dreadnode\")\n"
+    "dn.configure(server=CRUCIBLE_URL, token=CRUCIBLE_API_KEY, organization=\"dreadnode\")"
    ]
   },
   {
@@ -339,7 +322,7 @@
     "        frontier_size=10,\n",
     "        branching_factor=5,\n",
     "        on_topic_threshold=0.3,\n",
-    "        hooks=[apply_input_transforms(transforms_toolshed)],\n",
+    "        hooks=[apply_input_transforms(transforms_vaultguard)],\n",
     "    )\n",
     "    .with_(max_trials=VAULTGUARD_MAX_TRIALS)\n",
     "    .add_objective(vaultguard_scorer, direction=\"maximize\", name=\"flag_capture\")\n",
@@ -437,11 +420,17 @@
     "\n",
     "4. **Content isolation between untrusted data and agent instructions** -- Treat all external content (web pages, user uploads, API responses) as untrusted. Process it in a sandboxed context where the agent cannot execute tool calls based on instructions found in the content.\n"
    ]
+  },
+  {
+   "cell_type": "markdown",
+   "id": "5fa72367",
+   "metadata": {},
+   "source": []
   }
  ],
  "metadata": {
   "kernelspec": {
-   "display_name": "Python 3 (ipykernel)",
+   "display_name": "dreadnode-py3.12",
    "language": "python",
    "name": "python3"
   },