We Published a Formal Spec for Tamper-Evident AI Audit Chains

AIR Blackbox v1.2.6 ships a Claude Agent SDK trust layer, 4600 training examples, and the first published specification for HMAC-SHA256 audit chains in AI agent systems. The EU AI Act requires that high-risk AI systems automatically record events over their lifetime — and that those logs can't be quietly modified after the fact. Article 12 is specific: "High-risk AI systems shall technically allow for the automatic recording of events ('logs') over the lifetime of the system." Most teams building AI agents have no answer for this. They have print() statements. Maybe a log file. Nothing a regulator would accept. We just published a formal specification for how to solve this problem. It's open-source, it's free, and it's the first published spec of its kind. What we shipped in v1.2.6 Three things landed this week: 1. HMAC-SHA256 Audit Chain Specification v1.0 A formal, citeable spec that defines how to build tamper-evident audit trails for AI agents. Every record is linked to the previou