OpenTraining.ai adds on-device face redaction for smart-glasses footage

Open Technology says its OpenTraining.ai platform blurs faces, license plates and documents on the capture device before footage is stored, aiming to protect bystanders in smart-glasses recordings used to train humanoid robots. The company is also paying skilled workers for consented captures through its Training Economy program as Australia weighs new rules on wearable recording devices.

Why it matters: - OpenTraining.ai is trying to make smart-glasses capture more privacy-safe by redacting sensitive details before any footage is saved. - The approach targets a growing policy flashpoint in Australia, where regulators and lawmakers are scrutinizing wearable cameras. - The platform also ties worker compensation to dataset creation, which could influence how physical-AI training data is sourced.

What happened: - Open Technology LLC said its OpenTraining.ai capture system runs face detection and pixelation on the device in real time before footage is written to storage. - The stored file keeps the redacted image rather than the original camera feed. - The same pipeline also redacts licence plates and documents. - Unredacted frames are not uploaded to servers or viewed by human reviewers. - The platform records skilled trade work from a first-person view and packages the recordings as training data for humanoid robots and robot foundation models. - George Bancs, founder and CEO of Open Technology, said the system is designed to redact on the device before storage so that an unredacted copy is never created.

The details: - Each capture is consented at the point of recording and reviewed before it is added to a dataset. - Approved captures are stored as fixed versions with consent and provenance records attached. - The platform does not collect footage in the background or from scraped sources. - Through the Training Economy program, skilled workers are paid for each approved capture and receive a share of downstream dataset licensing revenue. - The program was featured on 7NEWS in October 2026. - Recordings document both the task being performed and related safety practices, including hazard identification. - Open Technology has authored hazard profiles across 16 industry verticals, starting with building and infrastructure. - The company is prioritising higher-risk work such as confined-space work and working at height. - Safe Work Australia reported 222 worker deaths from traumatic injuries in 2025, including 28 from falls from height. - OpenTechnology.ai currently integrates with Meta smart glasses, and native applications are in development.

Between the lines: - The announcement positions privacy controls as a product feature and a trust argument at a time when smart-glasses recording is drawing political attention. - The worker-payment model suggests Open Technology wants to frame dataset creation as labor, not just data extraction. - The focus on safety-critical trades gives the company a way to link training data collection with real-world risk reduction. - Bancs said tradespeople have always trained apprentices, and the Training Economy pays them when their expertise is used to train robots.

What's next: - Open Technology says native applications for its capture platform are in development. - The company is likely to keep expanding its safety-focused dataset library as it targets more high-risk industrial tasks. - The policy debate in Australia over wearable recording devices could shape how quickly similar systems are adopted.

The bottom line: - Open Technology is betting that on-device redaction and paid, consented captures can make smart-glasses training data acceptable for workers, regulators and customers alike.