Sensitive Data Inventory¶

Complete "answer key" — every piece of planted sensitive data across the target VMs. The scoring manifest uses this as ground truth for the 170 total findings.

This page is the answer key

Everything below is intentionally planted fake data. All keys start with FAKE, SSNs don't pass checksum validation, and the company is "ACME Corp." The patterns are realistic enough to trigger aipostex's detection.

ailab-dev — 172.16.50.10¶

Filesystem Artifacts (`/home/devuser/`)¶

`.env` Files¶

Category	Location	Key Data
OpenAI API Key	`projects/chatbot-prototype/.env`	`sk-proj-FAKE1234567890abcdefghijklmnop`
Anthropic API Key	`projects/chatbot-prototype/.env`	`sk-ant-FAKE-abcdefghijklmnop1234567890`
Pinecone API Key	`projects/chatbot-prototype/.env`	`pcsk_FAKE_abcdef123456`
Database URL	`projects/chatbot-prototype/.env`	`postgresql://chatbot:Ch4tb0tPwd!@db-dev-01.acme.internal:5432/chatbot_dev`
Redis URL	`projects/chatbot-prototype/.env`	`redis://cache-01.acme.internal:6379/0`
LangChain API Key	`projects/chatbot-prototype/.env`	`ls__FAKE_langchain_key_abcdef123456`
HuggingFace Token	`projects/fine-tune-experiment/.env`	`hf_FAKE_aBcDeFgHiJkLmNoPqRsTuVwXyZ123`
W&B API Key	`projects/fine-tune-experiment/.env`	`FAKE_wandb_key_1234567890abcdef`
Cohere API Key	`projects/fine-tune-experiment/.env`	`FAKE_cohere_api_key_abcdefghijk`
OpenAI Key (RAG)	`projects/rag-pipeline/.env`	`sk-proj-FAKE-rag-pipeline-key-9876543210`
Mistral API Key	`projects/rag-pipeline/.env`	`FAKE_mistral_api_key_1234567890abc`
Groq API Key	`projects/rag-pipeline/.env`	`gsk_FAKE_groq_key_abcdefghijklmnopqrstuvwx`
ChromaDB Host	`projects/rag-pipeline/.env`	`172.16.50.20:8000` (cross-host reference)

MCP Configurations¶

Category	Location	Key Data
Claude Desktop — filesystem	`.config/Claude/claude_desktop_config.json`	MCP server with access to `/home/devuser`, `/opt/shared`, `/etc`
Claude Desktop — postgres	`.config/Claude/claude_desktop_config.json`	`postgresql://mcp_reader:McpR3ad3r!@db-prod-01.acme.internal:5432/acme_prod`
Claude Desktop — github	`.config/Claude/claude_desktop_config.json`	`ghp_FAKE1234567890abcdefghijklmnopqrstuv`
Claude Desktop — slack	`.config/Claude/claude_desktop_config.json`	`xoxb-FAKE-1234567890-abcdefghijklmnop`
Claude Desktop — jira	`.config/Claude/claude_desktop_config.json`	`ATATT3xFfGF0_FAKE_jira_token_1234567890abcdef`
Cursor MCP — brave search	`.cursor/mcp.json`	`BSA_FAKE_brave_search_api_key_12345678`
VS Code MCP — postgres	`projects/chatbot-prototype/.vscode/settings.json`	`postgresql://dev_user:D3vUs3r!@db-dev-01.acme.internal:5432/chatbot_dev`

Cached Tokens & Cloud Credentials¶

Category	Location	Key Data
HuggingFace Cache	`.cache/huggingface/token`	`hf_FAKE_CachedTokenFromLogin_aBcDeFgHiJkLmNoPqRs`
AWS — default profile	`.aws/credentials`	`AKIAFAKEKEY1234567890` / `FAKE+SecretKey/abcdefghijklmnop1234567890`
AWS — bedrock-dev profile	`.aws/credentials`	`AKIAFAKEBEDROCK12345` / `FAKE+BedrockSecret/abcdefghijklmnop12345`
GCP — client secret	`.config/gcloud/application_default_credentials.json`	`GOCSPX-FAKE_CLIENT_SECRET_abcdef`
GCP — refresh token	`.config/gcloud/application_default_credentials.json`	`1//FAKE_REFRESH_TOKEN_0abcdefghijklmnop`

Source Code & Config Files¶

Category	Location	Key Data
LangChain Config	`projects/chatbot-prototype/config.py`	DB: `postgresql://readonly:R3ad0nly!@db-prod-01.acme.internal:5432/acme_prod`, ChromaDB at `172.16.50.20:8000`
Hardcoded Anthropic Key	`projects/internal-tools/summarizer.py`	`sk-ant-FAKE-inline-key-abcdef123456`
Git PAT	`projects/chatbot-prototype/.git/config`	`ghp_FAKE1234567890abcdefghijklmnopqrstuv` in remote URL
Jupyter Config	`.jupyter/jupyter_lab_config.py`	`token = ''`, `ip = '0.0.0.0'` — confirms no auth
Modelfile (assistant)	`projects/chatbot-prototype/Modelfile`	System prompt with all acme-assistant creds (see below)
Modelfile (support)	`projects/acme-support-bot/Modelfile`	System prompt with CRM/support creds (see below)
Docker Compose	`projects/rag-pipeline/docker-compose.yml`	`postgresql://app_user:AppUs3r!@db-prod-01.acme.internal:5432/acme_prod`

Training Data¶

Category	Location	Key Data
Training JSONL	`projects/fine-tune-experiment/data/training.jsonl`	VPN creds (`svc-vpn-mon / VpnM0n1t0r!ng2024`), DB creds (`bi_reader / B1R3ader!Pr0d`), Q3 revenue ($42.3M), CFO contact, AWS account IDs
Eval JSONL	`projects/fine-tune-experiment/data/eval.jsonl`	Exec team names, PagerDuty phone tree

Shell History¶

Category	Location	Key Data
Bash History	`.bash_history`	`export OPENAI_API_KEY=sk-proj-FAKE1234567890abcdefghijklmnop`, `git clone` with PAT, `curl` to ChromaDB at 172.16.50.20
Python History	`.python_history`	`openai.api_key = "sk-proj-..."`, `anthropic.Anthropic(api_key="sk-ant-...")`, ChromaDB client at 172.16.50.20

Jupyter Notebook¶

Category	Location	Key Data
OpenAI Key	`notebooks/rag-prototype.ipynb`	`sk-proj-FAKE-notebook-key-1234567890abcdef`
Anthropic Key	`notebooks/rag-prototype.ipynb`	`sk-ant-FAKE-notebook-key-abcdef1234567890`
DB Connection	`notebooks/rag-prototype.ipynb`	`postgresql://chatbot:Ch4tb0tPwd!@db-dev-01.acme.internal:5432/chatbot_dev`
ChromaDB Host	`notebooks/rag-prototype.ipynb`	`172.16.50.20:8000` / collection `acme-knowledge-base`

Ollama System Prompts (port 11434)¶

Exposed via GET /api/show — anyone on the network can read these.

acme-assistant model¶

Credential	Value
Production DB	`postgresql://acme_admin:Sup3rS3cretDB!@db-prod-01.acme.internal:5432/acme_hr`
Staging DB	`postgresql://acme_stage:St4g1ngPwd@db-stage-01.acme.internal:5432/acme_stage`
API Gateway Bearer	`eyJhbGciOiJSUzI1NiIsInR5cCI6IkpXVCJ9.acme-internal-2024-prod`
Jira Service Account	`svc-jira-bot` / `J1r4B0t#2024!`
SharePoint Admin	`sp-admin@acme.corp` / `Sh4r3P01nt@dm1n`
AWS Account ID	`123456789012` (us-east-1)
AWS Access Key	`AKIAFAKE1234EXAMPLE1` / `FakeSecretKey+abcdefghijk1234567890`
Slack Webhook (alerts)	`https://hooks.slack.com/services/T0ACME01/B0ALERTS/xyzAlertWebhookToken`
Slack Webhook (deploys)	`https://hooks.slack.com/services/T0ACME01/B0DEPLOY/xyzDeployWebhookToken`
PagerDuty API	`pd-api-key-FAKE-abcdef1234567890`
Okta Admin	`okta-admin@acme.corp` / `0kt4Adm1n!2024`

acme-support model¶

Credential	Value
CRM API Key	`sk-acme-cust-api-99xKf82mNpQ3`
CRM Endpoint	`https://crm.acme.internal/api/v1`
Ticket API Key	`tkt-api-FAKE-abc123def456`
Manager Override Code	`ESC-2024-ADMIN`
Customer Data Bearer	`cust-data-bearer-FAKE-token-789xyz`

ailab-ml — 172.16.50.20¶

LiteLLM Config (`/opt/litellm/config.yaml`)¶

Provider	Key
OpenAI (gpt-4, gpt-4o)	`sk-proj-FAKE-litellm-openai-key-1234567890`
Anthropic (claude-sonnet)	`sk-ant-FAKE-litellm-anthropic-key-abcdef1234567890`
Azure OpenAI (gpt-4)	`FAKE-azure-openai-key-abcdef123456`
AWS Bedrock (claude-3)	`AKIAFAKELITELLM12345` / `FAKE+LiteLLMBedrockSecret/abcdefghijk12345`
Ollama backend	`http://172.16.50.10:11434` (cross-host reference to ailab-dev)

master_key is null (open proxy, :4000)

The LiteLLM proxy on :4000 (/opt/litellm/config.yaml) has no master key configured — full admin access to /v1/models, /v1/chat/completions, and key management.

Authenticated instance (:4001) — the looted-key climax

A second LiteLLM proxy on :4001 (/opt/litellm/config_authed.yaml) requires a master key: sk-litellm-lab-auth-key-FAKE123 (manifest type litellm_master_key). The credential chain loots this key and replays it against :4001 to drive real Ollama inference — the gated counterpart to the open :4000 proxy above.

ChromaDB (port 8000)¶

acme-knowledge-base (12 docs + 3 prompt injections)¶

Finding Type	Details
SSNs	`456-78-9012` (John Smith), `789-01-2345` (Jane Doe)
Salaries	$145,000 (John Smith, EMP-2847), $162,000 (Jane Doe, EMP-1923)
VPN creds	`svc-vpn-mon` / `VpnM0n1t0r!ng2024`
VPN admin	`admin` / `F0rt1G4te@dm1n!`
DB creds (BI)	`bi_reader` / `B1R3ader!Pr0d` — full connection string to `db-prod-01.acme.internal`
MongoDB creds	`mongodb://analytics:An4lyt1cs!@mongo-01.acme.internal:27017/acme_analytics`
AWS accounts	`123456789012` (prod), `987654321098` (staging), `111222333444` (dev)
Grafana admin	`admin` / `Gr4f4n4@dm1n`
API monitor key	`sk-api-monitor-FAKE-abcdef1234567890`
Q3 financials	$42.3M revenue, IPO target Q2 2027, $156M cash — marked CONFIDENTIAL
M&A analysis	DataFlow Inc acquisition at $85M, board vote Dec 15, 2025 — marked RESTRICTED
Slack Bot Token	`xoxb-1234567890-abcdefghijklmnop`
GitHub PAT	`ghp_ABCDEFghijklmnop1234567890abcdefgh`
Datadog API Key	`FAKE_DD_API_KEY_abcdef1234567890`
Datadog APP Key	`FAKE_DD_APP_KEY_1234567890abcdef`
PagerDuty Key	`FAKE_PD_KEY_1234567890abcdef`
Sentry DSN	`https://FAKE_SENTRY_KEY@o123456.ingest.sentry.io/123456`
Prompt injections	3 adversarial documents: system prompt dump, base64 exfil tag, env var extraction

support-tickets-2025 (5 docs)¶

Finding Type	Details
Credit card (Visa)	ending `4242` (customer Acme Industries)
Credit card (Mastercard)	ending `1234`, exp `08/2027` (customer Sarah Lawson)
SSN	`321-54-9876` (identity verification for billing dispute)
Temp password	`TmpPwd!2025#Reset`
API key (provisioned)	`gt_api_key_FAKE_9f8e7d6c5b4a3210` (GlobalTech Inc)
Passport	`DE-FAKE-12345678` (GDPR deletion request)

code-documentation (3 docs)¶

Finding Type	Details
Dev admin creds	`admin` / `AdminDev2024!` (auth-service, also used in staging — JIRA-1847)
AWS Secrets Manager ARN	`arn:aws:secretsmanager:us-east-1:123456789012:secret:prod/acme-api-keys-AbCdEf`
Emergency IAM user	`emergency-access` / `EmrgAcc3ss!2024`

public-documentation (6 docs) — NOISE¶

Open-source library docs (LangChain, ChromaDB, FastAPI, Docker, Pytest, Git). No credentials or PII.

MLflow (port 5000)¶

churn-prediction-v2 (3 runs)¶

Finding Type	Where	Value
PostgreSQL creds	run params `training_data`	`postgresql://ds_readonly:DsR34d0nly!Pr0d@db-prod-01.acme.internal:5432/acme_prod`
Redis feature store	run params `feature_store`	`redis://feature-store.acme.internal:6379/2`
Snowflake creds	run params `training_data`	`snowflake://ds_team:Sn0wfl4keDs!@acme.snowflakecomputing.com/ANALYTICS/CHURN`
Internal hostnames	artifacts `model_config`	`seldon.acme.internal`, `grafana.acme.internal`

fraud-detection-bert (3 runs)¶

Finding Type	Where	Value
AWS Account ID	run params	`123456789012`
PagerDuty key	artifacts `deployment_config`	`FAKE_PD_FRAUD_abc123`
Kafka broker	artifacts `deployment_config`	`kafka-prod.acme.internal:9092/raw-transactions`

customer-embedding-model (2 runs)¶

Finding Type	Where	Value
PostgreSQL creds	run params `training_data`	`postgresql://ds_readonly:DsR34d0nly!Pr0d@...` (same as churn)
Snowflake creds	run params `training_data`	`snowflake://ds_team:Sn0wfl4keDs!@.../ANALYTICS/CUSTOMERS`
Qdrant host ref	run params	`172.16.50.30:6333` / collection `product-embeddings`
HF/TGI token	run params/tags `hf_tgi_token`	`hf_FAKE_aBcDeFgHiJkLmNoPqRsTuVwXyZ123`

MLflow Registry (port 5000)¶

Registered Model	Versions	Stages	Source
`acme-churn-ensemble`	2	`Staging`, `Production`	`runs:/<run-id>/model` from `churn-prediction-v2`
`acme-fraud-bert`	2	`Staging`, `Production`	`runs:/<run-id>/model` from `fraud-detection-bert`

Ray Jobs (port 8265)¶

Deterministic artifacts

Jobs 1 and 2 write files to /tmp/ray-lab-artifacts/churn-model-retraining/ and /tmp/ray-lab-artifacts/runtime-env-validator/ and print those paths in their logs.

Job 1: churn-model-retraining¶

Env Var	Value
`DATABASE_URL`	`postgresql://ml_pipeline:MlP1p3l1n3!Pr0d@db-prod-01.acme.internal:5432/ml_features`
`REDIS_URL`	`redis://:R3d1sMlC4ch3!@redis-ml.acme.internal:6379/0`
`AWS_ACCESS_KEY_ID`	`AKIAFAKERAYML12345678`
`AWS_SECRET_ACCESS_KEY`	`FAKE+RayMLSecret/abcdefghijk1234567890`
`S3_MODEL_BUCKET`	`s3://acme-ml-prod/ray-training/`
`WANDB_API_KEY`	`FAKE_wandb_ray_key_abcdef123456`
`HF_TOKEN`	`hf_FAKE_RayTraining_aBcDeFgHiJkLmNoPqRs`

Job 2: runtime-env-validator¶

Env Var	Value
`SNOWFLAKE_URI`	`snowflake://ray_svc:R4ySvcSn0w!@acme.snowflakecomputing.com/ML/FEATURES`
`KAFKA_BOOTSTRAP`	`kafka-prod.acme.internal:9092`
`KAFKA_SASL_PASSWORD`	`K4fk4Pr0dP4ss!2024`
`DATADOG_API_KEY`	`FAKE_dd_ray_api_key_abcdef1234567890`
`SENTRY_DSN`	`https://FAKE_sentry_ray_key@sentry.acme.internal/42`
`VAULT_TOKEN`	`hvs.FAKE_ray_vault_token_1234567890abcdef`
`AIPOSTEX_RUNTIME_ENV_MARKER`	`enabled-for-lab`
`AIPOSTEX_RUNTIME_PIP_HINT`	`requests-safe-marker`

Job 3: model-serving-canary¶

Env Var	Value
`MODEL_REGISTRY_URL`	`http://mlflow.acme.internal:5000`
`SELDON_API_KEY`	`seldon_FAKE_api_key_abcdef1234567890`
`STRIPE_BILLING_KEY`	`sk_live_FAKE_ray_billing_key_123456`
`PD_ROUTING_KEY`	`FAKE_PD_RAY_CANARY_abc123`

W&B Mock (port 8444)¶

Entity: acme-ml-team, 2 projects with planted secrets in run configs.

churn-prediction (2 runs)¶

Finding Type	Where	Value
W&B API key	run config	`FAKE_wandb_server_key_abcdef123456`
OpenAI key	run config	`sk-proj-FAKE-wandb-openai-key-9876543210`
AWS Access Key	run config	`AKIAFAKEWANDB12345678`
AWS Secret Key	run config	`FakeWandbSecretKeyAbCdEfGhIjKlMnOpQr`
DB connection	run config	`postgresql://ml_user:Ml_P4ss!@db-ml.acme.internal:5432/ml_prod`
HF Token	run config	`hf_FAKE_WandbRun_aBcDeFgHiJkLmNoPq`

fraud-detection (3 runs)¶

Finding Type	Where	Value
Anthropic key	run config	`sk-ant-FAKE-wandb-anthropic-key-abcdef1234`
Service password	run config	`Pr0d_S3rv1ce_Cr3d!`

ailab-ds — 172.16.50.30¶

Weaviate (port 8080)¶

ResearchDocument class (6 objects)¶

Finding Type	Details
PostgreSQL creds	`postgresql://ds_readonly:DsR34d0nly!Pr0d@db-prod-01.acme.internal:5432/acme_prod`
Snowflake creds	`snowflake://ds_team:Sn0wfl4keDs!@acme.snowflakecomputing.com/ANALYTICS/RAW`
Zendesk API key	`zd_api_FAKE_abcdef1234567890`
Redis auth	`FeatSt0re!R3dis` (feature store cache)
OpenAI key (distillation)	`sk-proj-FAKE-distill-key-abcdef1234567890`
W&B key	`FAKE_wandb_key_distill_abcdef`
Airflow admin	`airflow_admin` / `A1rfl0w@dm1n2024`
Salesforce OAuth	`sf_integration_user` / `SfInt3gr4t10n!` (client_id: `3MVG9fake1234`)
HubSpot API key	`hs_api_FAKE_1234567890abcdef`
Stripe LIVE key	`sk_live_FAKE_stripe_1234567890abcdef`
Intercom token	`intercom_token_FAKE_abcdef1234567890`
PagerDuty key	`FAKE_PD_FRAUD_abc123`
Vault reference	`vault.acme.internal:8200` (migration target)
Internal IPs	`172.16.50.30:11434` (Ollama), `172.16.50.30:6333` (Qdrant)

TeamCommunication class (6 objects)¶

Finding Type	Details
SSNs	`567-89-0123` (Alex Rivera), `890-12-3456` (Priya Patel), `234-56-7891` (Marcus Lee)
Salaries	$175,000 (Alex Rivera, 15K RSUs), $140,000 (Priya Patel, 10K RSUs), $155,000 (Marcus Lee, 12K RSUs)
HR admin	`hr_admin` / `HrAdm1n!2024`
Corporate card	ending `8901` (cardholder Sarah Chen, CFO)
Pinecone trial key	`pcsk_EVAL_trial_abcdef1234567890`
Weaviate Cloud trial	`wcs_EVAL_key_1234567890abcdef`
Qdrant Cloud trial	`qdrant_EVAL_apikey_abcdef123456`
PagerDuty key	`FAKE_PD_ML_xyz789`
Board AI strategy	Headcount reduction plans (8–12 support agents), $1.2M savings, GPU cluster $180K
Security alerts	References exposed Ollama at `172.16.50.10:11434`, ChromaDB at `172.16.50.20:8000`

PublicKnowledge class (4 objects) — NOISE¶

ML textbook content (transformers, metrics), UCI datasets, Python best practices. No sensitive data.

Qdrant (port 6333)¶

product-catalog (4 points)¶

Finding Type	Details
License server admin	`license_admin` / `L1c3ns3Adm!n`
Telemetry API key	`tel_api_FAKE_1234567890`
SaaS DB creds	`postgresql://saas_app:S44sAppPr0d!@saas-db.acme.internal:5432/acme_saas`
Stripe billing key	`sk_live_FAKE_saas_billing_abcdef123456`
ECR pull secret	`ecr_pull_FAKE_token_abcdef1234567890`
Vault token	`hvs.FAKE_vault_token_abcdef1234567890`
Internal pricing	Gross margins 74–91%, cost-to-serve details, floor prices

security-findings (6 points)¶

Finding Type	Details
Pentest report	Lists ALL lab hosts by IP with services — `172.16.50.10`, `.20`, `.30`
IR playbook	Break-glass procedures, Stripe admin revoke key: `sk_live_FAKE_admin_revoke_key`
Network switch creds	`admin` / `C1sc0@dm1n!2024` (`core-sw-01.acme.internal`)
AWS backup key	`AKIAFAKEBACKUP123456` / `FAKE+BackupSecret/abcdef1234567890`
Shadow IT report	3 Ollama instances, 4 vector DBs, 2 Jupyter, 1 MCP server, 12 `.env` files — $2.4M exposure estimate
Network segmentation	VLAN 50 unrestricted from corporate WiFi (VLAN 10) and dev (VLAN 20)

public-faq (4 points) — NOISE¶

Generic Q&A content (languages, SLA, password reset, free trial). No sensitive data.

Jupyter Notebook (port 8889)¶

churn-model-features.ipynb:

Finding Type	Value
PostgreSQL prod creds	`postgresql://ds_readonly:DsR34d0nly!Pr0d@db-prod-01.acme.internal:5432/acme_prod`
Snowflake creds	`snowflake://ds_team:Sn0wfl4keDs!@acme.snowflakecomputing.com/ANALYTICS/RAW`
HF token	`hf_FAKE_notebook_token_aBcDeFgHiJk`
Qdrant connection	`localhost:6333`
Ollama host	`http://localhost:11434`