PoC V2 fully activated, BLS accelerated by 200x, security vulnerabilities closed, network transitioned to Qwen3-235B.
Proof of Computation V2 has progressed from tracking mode (v0.2.8) to full enforcement (v0.2.9). Now every node is required to prove that it genuinely performs AI computations. PoC V2 is integrated directly into vLLM — switching between inference and verification occurs instantly, without model reloading.
GitHub ↗BLS signature cryptographic verification has been optimized by 200x thanks to the blst library and precomputation of slot keys. A critical operation performed on every block is now almost instantaneous.
GitHub ↗SSRF vulnerabilities (redirection of requests to internal services), vote flipping (overwriting PoC validations), integer overflow (overflow during escrow calculation) have been fixed, and DoS protection has been added by limiting request body sizes.
GitHub ↗All other models have been removed from governance. Consolidation simplifies validation and improves reliability — all nodes are guaranteed to work with the same model.
GitHub ↗Analysis showed that 24 addresses generated PoC batches but did not participate in validation. These nodes have been removed from the allowlist. A Transfer Agent whitelist has been added for additional inference access control.
GitHub ↗