Be careful robot companies Under U.S. law, a company that trains models on data collected unlawfully or without consent can still be liable — even if they did not record it. @eddybuild can you provide proof that you own the dataset and that all 2,153 workers gave proper legal consent before these videos were distributed? 1. Signed consent forms from all workers (2,153 individuals) Each worker must sign: •video + audio release form •dataset usage consent •consent for public redistribution •biometric data consent (hands, faces, gait = biometric identifiers) Without explicit written consent → illegal under multiple U.S. state laws. ⸻ 2. Written authorization from each factory Every factory where recording occurred must provide: •site filming authorization •permission to publicly release footage •confirmation no trade secrets are exposed •IP transfer or licensing agreement Factories own: •their workflow •machinery layout •operational processes This is protected IP. ⸻ 3. Proof he owns the copyright or the license He must show: •who recorded the videos •who funded the recording •who owns the cameras •who owns the footage under employment law •IP assignment or release agreements If workers recorded the videos as part of employment, the factory/employer owns the footage — not him. ⸻ 4. Data protection assessment (especially biometrics) U.S. states like: •Illinois (BIPA) •Texas •California (CCPA/CPRA) require strong justification + signed consent before collecting or distributing biometric data. Factory footage almost always contains: •faces •gait •body measurements •hand motion patterns All are “biometric identifiers.” ⸻ 5. Removal of personally identifiable information If any worker is identifiable from: •face •voice •name tags •tattoos •workspace arrangement •conversations Then he must: •censor •blur •anonymize before public release. ⸻ 6. Documented chain of custody He needs records showing: •how the footage was stored •when it was transferred •who handled it •security protections •deletion of originals from unauthorized parties This matters for privacy + trade secret compliance. ⸻ 7. Legal review by counsel For a dataset involving: •human subjects •workplace environments •global factories A licensed attorney must provide a written clearance. Without legal review → massive risk. ⸻ 8. Export law compliance If recordings came from: •China •India •Southeast Asia •Latin America He must ensure: •data export was legal in that country •workers were paid or compensated according to local law •no labor laws were violated Otherwise, cross-border data transfer becomes illegal. ⸻ 9. Licensing terms that forbid misuse He must publish: •a dataset license •limitations on commercial use •privacy guarantees •the right for participants to request deletion This is required for ethical AI datasets.