Snyk Vol II: AI's real footprint is 3x model counts

Snyk has published Volume II of the State of Agentic AI Adoption: AI-BOM telemetry from 3,044 enterprise accounts across the Americas, EMEA and Asia-Pacific — approximately 1.39 million code repositories analyzed in June 2026. The structural thesis from Volume I's 500+ environments — AI is no longer deployed as isolated features but as interconnected execution systems with their own supply chains, dependencies and emerging autonomous behavior — has now been validated twice, across three regions.

Four counterintuitive numbers

Capability: conservative deployments, frontier already in production

Volume II introduces capability for the first time, scoring deployed models with Epoch AI's ECI (frontier 160 as of April 2026). The weighted average of deployed proprietary models is 135.4 — about 25 points below the frontier, dragged down by a long tail of older models (gpt-3.5-turbo, claude-3-opus) kept for cost or stability. But frontier-adjacent deployments (155-159 ECI) were observed 1,874 times, all proprietary — the leading edge is already in production. Open-weight models trail the closed frontier by about 4 months / 8 ECI points. The takeaway: the local/open-versus-frontier tradeoff is becoming a core architecture decision, and governance should be tiered by capability, not just count.

The lineage gap is the governance gap

The report's most uncomfortable number: only 50.8% of model-bearing accounts declare any dataset in their repositories, at about 0.36 datasets per model. Roughly half of model-deploying organizations have no visible code-level link to the data that shaped their models — in every region. Bias assessment, regulatory compliance, audit response, incident investigation and IP assurance all depend on knowing what data shaped a model's behavior. The report draws the software analogy plainly: enterprise AI is evolving along the same path as every major infrastructure transition — from isolated experimentation, to operational dependence, to systemic criticality. The difference is that these systems are becoming autonomous while the transition is still unfolding.

Supply chain: external dependencies dominate

77.4% of AI tools come from third-party packages, about 3.4:1 over custom tools. AI supply chain security is software supply chain security with new, behavioral dimensions. For governance leaders this means model-focused governance covers about a third of the real footprint — datasets, tools, frameworks and third-party packages all contribute meaningfully to the risk profile.

What it means for enterprises

The bottom line

The report's value is not in any single number but in the ratios: models are a third of the real AI footprint; half of deployers cannot trace their data; frontier models are already running in production. Enterprise AI governance is becoming a distinct operational discipline — adjacent to but different from application security, cloud governance and software supply chain. Its object is interconnected autonomous systems whose behavior emerges from models, tools, datasets, orchestration and external services. Govern by model inventory alone, and you are managing the tip of the iceberg.

References

• Snyk: 2026 State of Agentic AI Adoption — Volume II (2026-08) — https://res.cloudinary.com/snyk/image/upload/v1785759343/Volume_II-2026_The_State_of_Agentic_AI_Adoption_August_2026_a9jyao.pdf