Skip to content

Security: Pika-Labs/Pika-Experiments

Security

SECURITY.md

Security policy

Reporting a vulnerability

If you discover a security issue in Pika Experiments — credential exposure, an unsafe default in any experiment folder, a way to escape the localhost sandbox, anything that could harm someone running this code — please email support@pika.art with subject line starting [security] rather than opening a public issue or PR.

We aim to:

  • Acknowledge within 2 business days
  • Provide a fix or mitigation timeline within 7 days for high-severity issues
  • Coordinate public disclosure once a fix is shipped

Threat model for this repo

This repo ships experimental prototypes, not production software. Each subfolder is a small standalone app intended for local development on a trusted machine. The threat model assumes:

  • The user runs the experiment on localhost or behind their own access controls.
  • The user provides their own provider credentials (OpenAI, Google, etc.) via local environment files.
  • Tokens and refresh tokens live in local files (.env, .pika-token.json, .google-token.json) that are git-ignored.

The experiments are not hardened for public hosting. Exposing them to the open internet without your own auth, rate limiting, and isolation is out of scope and unsafe.

What's in scope

Issues we want to know about:

  • Hardcoded secrets, tokens, or PII committed to this repository
  • Defaults that would cause a user's credentials to leak (e.g., logging an API key, writing a token to a world-readable path)
  • Path traversal, SSRF, or command injection in any experiment's server code
  • An experiment instructing the agent in a way that could exfiltrate user data through the MCP server
  • CORS / auth misconfigurations that would let a third-party site take over a running local instance

What's out of scope (forwarded upstream)

  • Issues with the upstream Pika MCP server (mcp.pika.me) → support@pika.art with subject MCP server: ...
  • Provider-side issues (OpenAI Realtime, Google APIs) → those providers directly
  • "I exposed this to the internet without auth and got pwned" → that's a usage issue, see the threat model above
  • Bugs that don't have a security impact → open a normal GitHub issue

Disclosure

Once a fix lands, we'll credit the reporter (if desired) in the commit message and any related advisory. If the issue affects an upstream repo (Pika-Plugins, Pika-Skills, or pika-mcp-server), we'll coordinate disclosure across the relevant repos.

Last updated: 2026-05-25.

There aren't any published security advisories