TRACCIA: AN OPENTELEMETRY-BASED GOVERNANCE PLATFORM FOR AI SYSTEMS

Modern AI systems, particularly those powered by Large Language Models (LLMs) and autonomous agents, operate in non-deterministic, unbounded state spaces. Traditional software assurance methods—such as static analysis and unit testing—are insufficient for these dynamic environments. While current observability tools can monitor system metrics, they often fail to bridge the gap between raw technical data and the prescriptive requirements of legal frameworks like the EU AI Act. This creates a "last mile" problem where organizations struggle to prove compliance for their AI deployments.

Traccia addresses this by integrating AI governance directly into the execution path using OpenTelemetry (OTel). Rather than treating observability as a separate, descriptive task, Traccia functions as a prescriptive governance layer. It captures telemetry data—including prompts, token usage, tool invocations, and safety triggers—and transforms them into structured, policy-compliant artifacts. By extending OTel semantic conventions, the platform creates an immutable, hashed trace ledger that provides a clear chain of custody for AI decision-making.

Traccia automates the creation of compliance evidence packages by appending tamper-resistant fingerprints and SHA-256 content hashes to execution traces. This allows enterprises to map specific runtime behaviors directly to regulatory requirements, such as Articles 12, 14, 19, 26(6), and 50 of the EU AI Act. By maintaining this cryptographic link between high-level policy and low-level execution, Traccia enables continuous, evidence-based auditing without compromising data privacy.