DEPA's Aadhaar-style framework for AI training is doomed to fail, critic argues
kingslyj · x · 2026-11-04
A critic argues that anyone familiar with how India's Account Aggregators actually work knows DEPA is a bad framework for users, and its push into AI will be worse. An accompanying diagram lays out the proposed DEPA-for-AI-training architecture: data providers anonymize datasets, training happens inside secure data environments (zero-trust access, differential privacy, audit logs), followed by fairness/security audits before pilot deployment.
The author says this replicates the Aadhaar playbook for AI and warns that the renewed push for "formal proofs + AI" repeats the same mistakes, predicting DEPA will fail just as Aadhaar did.
More from Safety
- AI is forcing us to pay off 40 years of technical debt — pseudolus · 2026-11-04
- AI removes the competence barrier that kept national-scale cyberchaos in check — harris_edouard · 2026-11-04
- Exit polls: AI ranked last among voter priorities in 2026 midterms — The Decoder · 2026-11-04
- Tsinghua's BEMULA benchmark: no LLM agent memory system passes over 1.5% of forget requests — rohanpaul_ai · 2026-11-04
- UK AI minister warns serious cyberattack is "extremely likely" and country unprepared — Polymarket · 2026-11-04
- Gemini trains on chat history even for paying subscribers, no opt-out — jameslk · 2026-11-04