Lebrel opens beta: encrypted, uncensored frontier models with on-chain verifiable answers
HeyAmit_ · x · 2026-11-05
Lebrel opened a beta for encrypted, verifiable, uncensored serving of frontier open-weight models — 4 free hours daily (16:00–20:00 Madrid). Its "Proof of Edition" open standard includes:
- No added refusal layer, prompts not stored
- Signed receipts on every answer naming the exact model, precision, and engine, verifiable in-browser and registered on Solana
- Requests decrypted only inside attested hardware — neither Lebrel nor the cloud provider can read them
- Open weights published with per-file hashes
The motivation: resellers have been passing off Claude's answers as their own and silently swapping models. SDKs are open source (pip/npm).
More from Safety
- Safety researcher Heidy Khlaaf challenges AI labs: when they say "aligned AI," ask with whom — sebkrier · 2026-11-06
- Google paper: 5 query-echoing docs (0.1% of corpus) gut RAG retrieval — rohanpaul_ai · 2026-11-06
- AI safety researcher David Krueger calls for immediate indefinite moratorium on frontier AI — DavidSKrueger · 2026-11-06
- OpenAI's cyber model found 948 Linux vulns and 111 Firefox RCEs, says Greg Brockman — gdb · 2026-11-06
- Absurd flaw: any text over 256 chars reportedly works as a root password — generativist · 2026-11-06
- Independent audit of OpenAI's Astra safety report finds key data withheld, causal claim shaky — kevrussell · 2026-11-06