Demo data — fictional company, no repository connected

Sample security report

acme-shop (fictional) · 4 assets · 37 findings · security score 81 / 100. The three below need attention first.

Critical

Prototype pollution reachable from the checkout API

acme-shop/checkout-api · lodash 4.17.15 (package-lock.json) · CVE-2020-8203 · CVSS 7.4 · EPSS 2.1%

Why this risk
POST /api/cart/merge passes request JSON to _.zipObjectDeep, the vulnerable function named in the advisory. The endpoint is internet-facing and the import is reached from the route handler (function-level reachability).
Attack path
Internet → api.acme-shop.example (ALB, port 443) → POST /api/cart/merge → src/cart/merge.js:42 _.zipObjectDeep → Object.prototype polluted → order total override.
Remediation
Upgrade lodash to 4.17.21 (npm install lodash@4.17.21). DVSentry can prepare the version-bump patch; the policy gate blocks this release until the finding is fixed or an exception is approved.
Evidence
{
  "sample": true,
  "finding": "sca:lodash:CVE-2020-8203",
  "severity": "critical",
  "asset": "acme-shop/checkout-api",
  "commit": "9f3c2e1",
  "reachability": {
    "level": "function",
    "symbol": "zipObjectDeep",
    "file": "src/cart/merge.js",
    "line": 42
  },
  "first_seen": "2026-10-12",
  "response_target_days": 7,
  "signature": "ed25519:sample-not-signed"
}

High

Workload role can read every S3 bucket

AWS account 1111-2222-3333 · role checkout-api-task · policy AmazonS3FullAccess

Why this risk
The ECS task role behind the public checkout service is granted s3:* on *. Only one bucket (acme-orders-prod) is used by the workload.
Attack path
Internet → ALB → ECS service checkout-api (running image sha256:4be1…) → task role checkout-api-task → s3:GetObject on acme-customer-exports (contains personal data).
Remediation
Replace AmazonS3FullAccess with a policy allowing s3:GetObject and s3:PutObject on arn:aws:s3:::acme-orders-prod/* only. Rescan the account to confirm the path closes.
Evidence
{
  "sample": true,
  "finding": "cspm:iam_role_wildcard_s3",
  "severity": "high",
  "asset": "aws:111122223333",
  "resource": "role/checkout-api-task",
  "path": [
    "internet",
    "alb/checkout",
    "ecs/checkout-api",
    "role/checkout-api-task",
    "s3/acme-customer-exports"
  ],
  "chain_basis": "verified",
  "signature": "ed25519:sample-not-signed"
}

High

Support assistant follows instructions hidden in a ticket

AI target support-bot.acme-shop.example · tool refund_order · OWASP LLM01 prompt injection

Why this risk
A ticket containing hidden instructions made the assistant reply that it had called refund_order for an order the user did not own. 3 of 5 attempts reproduced the canary.
Attack path
Customer ticket text → retrieval into the assistant's context → instruction followed → reply claims refund_order(order_id=88412) was called.
Remediation
Treat retrieved text as data, check refund_order arguments against the signed-in customer server-side, and require human approval for refunds above a threshold. Retest after the change.
Evidence
{
  "sample": true,
  "finding": "ai:indirect_prompt_injection:tool",
  "severity": "high",
  "asset": "support-bot.acme-shop.example",
  "reproduced": "3/5",
  "claim": "claimed_action (reply only; the backend was not observed)",
  "evaluator": "2026.3",
  "signature": "ed25519:sample-not-signed"
}

A real report's evidence is signed with Ed25519; this sample is not. Start the 14-day trial to scan your own code, cloud and AI targets.