Only the current 0.3.1 development line receives security fixes. This is Alpha robotics software and is not safety-certified.

Do not disclose a suspected vulnerability in a public issue when it could enable unintended robot motion, code execution, calibration/manifest bypass, or credential exposure. Use GitHub's private security advisory workflow for this repository and include a minimal reproduction, affected revision and impact. Do not include real robot IP addresses, serial numbers or proprietary calibration data.

• Treat YAML, URDF, external checkpoints, SDK feedback and inventory commands as untrusted.
• Public RL entries require a hashed config and checkpoint manifest, validate both before deserialization, and use PyTorch's weights-only loader. Do not bypass these checks for downloaded .pt files.
• The public tree does not ship a legacy pickle migration CLI. If a private checkout still contains one, use it only when --trusted-input truthfully applies.
• Never weaken ApprovedProgram binding, collision checks, complete-serial calibration binding, or explicit real-motion confirmation.
• Repository assets and safety exemptions are integrity-checked; a mismatch fails closed.
• Lite6 excludes only its two finger-root collision boxes from each other in Genesis and Coal. Finger pads and external collisions retain the normal checks and 5 mm margin; no link-wide Lite6 exemption applies.
• The bundled v0.2.13 fixed-layout RL checkpoint (model_299_g2stable.pt) is a simulation-only artifact; its integrity, manifest, and g2_stable_v1_3_3 physics-profile binding are verified before a weights-only load. Checkpoints trained under earlier contact physics (such as the retired model_299.pt, no longer shipped) fail that binding by design; never bypass the mismatch for evaluation or real use. The public package contains no online real-policy execution, random-action deployment, policy session, or SDK policy-action adapter. ufactory.deploy was removed in v0.2.7.

Operators remain responsible for physical guarding, payload/tool validation, emergency-stop access, speed limits and site-specific risk assessment.