Why AGI Cannot Emerge Without Private Data

The public data ceiling

Foundation models have achieved remarkable capabilities by training on public web data. But public data represents a narrow slice of useful knowledge.

The most valuable information in the world — proprietary market research, internal operational procedures, clinical trial results, legal precedents, financial models, engineering specifications — lives behind organizational walls. It is not on the open web. It is not in Common Crawl. It is not in anyone’s training set.

According to IBM’s 2025 CEO Study, 72% of CEOs view proprietary data as key to unlocking the value of generative AI. The study, conducted with Oxford Economics across 2,000 CEOs in 33 countries, also found that 61% were actively adopting AI agents and preparing to implement them at scale.

This signals a fundamental truth: the next leap in AI capability is not about bigger models or better architectures alone. It is about access to private data.

Why agents change the equation

The shift from chatbots to agents changes what data access means in practice.

A chatbot can work with vague answers. An agent executing a workflow cannot. When an agent is processing a compliance check, analyzing a contract, or making a procurement decision, it needs specific, verified, current information from proprietary sources.

PwC’s 2025 AI Agent Survey found that 79% of US business executives say AI agents are already being adopted in their companies, with 88% planning to increase AI-related budgets in the next 12 months. Among adopters, 66% said agents were already delivering measurable productivity value.

But the same research highlights a gap: enterprises have agents, but agents lack access to the private data they need to operate reliably.

The trust and quality problem

Without verified private data, agents face three failure modes.

Hallucination. Agents generate plausible but incorrect information when they lack grounding in authoritative sources. In regulated industries — financial services, healthcare, legal — this is not just embarrassing, it’s dangerous.

Generic outputs. An agent advising on a specific contract cannot rely on general legal knowledge. It needs the actual contract, the relevant regulatory context, and the organization’s internal policies. Public data cannot provide this.

Missing context. Enterprise decisions depend on institutional knowledge that exists only in internal documents, databases, and knowledge bases. An agent without access to this context is fundamentally limited in what it can accomplish.

The infrastructure gap

The problem is not that organizations lack valuable data. It is that no infrastructure layer exists to make that data safely, reliably, and economically available to AI agents.

IBM’s study also found that 50% of executives said rapid AI investment had left them with disconnected technology. The data is there. The agents are there. The connective tissue between them is missing.

Deloitte’s 2026 State of AI in the Enterprise report identified search and knowledge management as among the most impactful generative AI areas. The report noted that agentic AI is expected to have high impact in knowledge management alongside customer support, supply chain, R&D, and cybersecurity.

What is needed is a purpose-built infrastructure layer that:

• Transforms private data into agent-consumable retrieval units
• Provides structured facts, entities, and metadata — not just text chunks
• Enforces granular access controls at the tenant, user, and agent level
• Tracks provenance so every retrieval is auditable
• Enables pricing and monetization for data owners
• Delivers results at agent-grade latency

The path forward

The path to more capable AI — and ultimately to AGI — runs through private data. Not by hoarding it into massive training sets, but by creating the infrastructure layer that allows agents to retrieve it safely, cite it accurately, and compensate data owners fairly.

KPMG’s AI governance framework for the agentic era specifically emphasizes traceable inter-agent handoffs, explainability, confidence thresholds, guardrails, and human oversight, alongside strict access controls and privacy protections. These are not afterthoughts — they are prerequisites.

Gartner predicts that 40% of enterprise applications will feature task-specific AI agents by the end of 2026, up from less than 5% in 2025. The firm further forecasts that agentic AI could drive approximately 30% of enterprise application software revenue by 2035, exceeding $450 billion.

The agents are coming. The data they need is private. The infrastructure to connect them safely doesn’t exist yet. Building it is the most important unsolved problem in the path toward genuinely capable autonomous systems.

Key takeaways

• Foundation models hit a quality ceiling without access to private enterprise data
• 72% of CEOs identify proprietary data as key to unlocking AI value (IBM 2025)
• AI agents need structured, verified private data — not just text chunks from the web
• The infrastructure layer for private data retrieval, trust, and monetization is the critical missing piece
• The market for this infrastructure is growing rapidly: 40% of enterprise apps will have agents by 2026 (Gartner)