Standard Compute Vulnerability Disclosure

generated: '2026-09-02'
method: searched
probe: true
source: https://standardcompute.com/security
program:
  published: true
  type: responsible disclosure (no bug bounty)
  bounty: false
  bounty_platform: null
  contact_email: contact@standardcompute.com
  submission_channel: email
  required_report_contents:
    - detailed description of the vulnerability
    - steps to reproduce
    - supporting evidence
  response_commitments:
    acknowledgement: within 2 business days
    initial_assessment: within 7 business days
    ongoing: keeps the reporter informed of remediation progress
  safe_harbor:
    offered: true
    text: >-
      "We will not take legal action against researchers who report vulnerabilities
      in good faith and do not access other users' data or disrupt the service."
    conditions:
      - good-faith reporting
      - no access to other users' data
      - no disruption of the service
security_txt:
  served: false
  probed:
    - url: https://standardcompute.com/.well-known/security.txt
      status: 404
    - url: https://api.stdcmpt.com/.well-known/security.txt
      status: 404
  note: >-
    The policy exists and meets RFC 9116's substance — a contact, a policy page,
    stated response times and safe harbour — but the machine-readable file is not
    served. Publishing /.well-known/security.txt with Contact:
    mailto:contact@standardcompute.com and Policy:
    https://standardcompute.com/security would be a five-minute fix.
published_security_practices:
  api_key_encryption: AES-256-GCM at rest, per-account derived key stored separately
  key_visibility: plaintext shown once at creation; only an encrypted blob plus last four characters retained
  key_recovery: none — provider states no administrative path to recover plaintext
  request_authentication: HMAC-SHA256 with constant-time comparison
  brute_force_control: failed authentication attempts rate limited
  transport: TLS 1.2+, Mozilla Intermediate profile, plaintext HTTP refused (not redirected)
  database_isolation: row-level security policies enforced in the database engine, tested by cross-account integration tests
  prompt_logging: >-
    None. Prompt and response content is not logged, stored or inspected on any
    plan, and the provider states there is no analytics or debugging mode that
    enables it. Only metadata — timestamp, model identifier, token count — is
    retained, for usage tracking and fair-use enforcement.
  upstream_isolation: >-
    Per-provider TLS with dedicated service credentials; credentials for one upstream
    cannot access another. The routing pool spans multiple providers and countries,
    so the jurisdiction handling a request can vary.
  training_on_customer_data: >-
    Standard Compute does not train its own models on customer prompts or outputs.
    It explicitly declines to warrant upstream providers' training policies and tells
    compliance-bound buyers to read the upstream provider's own terms.
evidence:
  - source: https://standardcompute.com/security
    status: 200
    fetched: '2026-09-02'
    kind: disclosure page
    keywords:
      - vulnerability
      - responsible disclosure
      - safe harbor