$39M+ in Fines for
Unauditable AI Decisions
The FTC has established algorithmic disgorgement — if you build AI models from improperly collected data, you must delete the models themselves. Courts are rejecting 'human in the loop' defenses as purely symbolic. The EU AI Act enforcement begins 2025-2026.
Featured Cases
Amazon / Alexa
May 2023
Retained children's voice recordings indefinitely, even when parents requested deletion. Children's speech patterns provided unique training dataset. Prohibited from using retained children's voice data to train algorithms.
Source: FTC Press ReleaseAmazon / Ring
May 2023
Employees had unfettered access to customer security camera videos, some viewed intimate recordings. Used customer videos to train algorithms without consent. Ordered to delete all models derived from improperly accessed videos.
Source: FTC Press ReleaseWW International / Kurbo
March 2022
Kurbo weight management app collected children's data without parental consent. First COPPA case requiring algorithmic disgorgement — all models built from children's data ordered destroyed.
Source: FTC Press ReleaseUber / Ola — Robo-Firing
April 2023
Drivers algorithmically deactivated based on automated fraud scores without meaningful human review. Court rejected 'human in the loop' defense — review was 'purely symbolic.' Ruled that trade secrets cannot justify withholding info about algorithmic decisions.
Source: Fountain CourtiTutorGroup
August 2023
EEOC's first AI discrimination settlement. Hiring algorithm automatically rejected female applicants over 55 and male applicants over 60. Discovered when applicant submitted identical applications with different birth dates — the younger one got the interview.
Source: Sullivan & CromwellWhat went wrong in every case
The same accountability gaps appear across facial recognition, voice assistants, hiring algorithms, and gig economy platforms.
- AI decisions had no verifiable audit trail — "human in the loop" was purely symbolic
- Training data provenance unprovable — improperly collected data used to train models
- Algorithmic disgorgement is the new penalty — FTC orders deletion of models, not just data
- Age discrimination baked into hiring algorithms — automatically rejected older applicants
- EU AI Act enforcement begins 2025-2026 — up to $38M or 7% of global turnover
How tamper-proof evidence changes the equation
Cryptographic audit trails make it mathematically impossible to alter or destroy records without detection.
Every AI Decision is Recorded
Cryptographic receipts capture each algorithmic decision — prompts, model configs, inputs, outputs — at the moment it occurs. When regulators ask 'prove your AI was monitored,' the answer is a verifiable chain, not a policy document.
Data Provenance is Cryptographic
From data collection through model training, every step has a cryptographic receipt. When the FTC asks where training data came from and whether consent was obtained, the chain provides mathematical proof.
Fairness Metrics Are Tamper-Proof
Bias audits and fairness metrics are captured with cryptographic timestamps. Organizations can prove that monitoring was continuous, not just performed at audit time — meeting both FTC and EU AI Act requirements.
Ready to make your audit trail tamper-proof?
Create your free account and start protecting your data with cryptographic evidence that can't be altered or destroyed.