OpenObscure launches on-device AI privacy firewall

OpenObscure is interesting because it treats privacy as an in-process control plane, not a post-hoc policy layer. If the setup really is as light as the pitch, this has a better shot at sticking than most agent-security tools.

• –FF1 format-preserving encryption is the right call for cards, SSNs, and other structured PII because the model still sees realistic input.
• –Localhost-only, no-telemetry, and no default keys make this feel like actual security infrastructure, not a SaaS wrapper.
• –The scope is broader than a proxy: text, images, audio, and tool results are all covered, plus mobile SDKs and OpenClaw integration.
• –The EU AI Act / cognitive firewall angle is interesting, but it still needs real-world validation before anyone reads it as compliance coverage.
• –The market is already crowded with redaction tools, cloud gateways, browser extensions, and enterprise DLP, so OpenObscure wins only if setup stays frictionless.