Dun & Bradstreet will integrate its Commercial Graph—a comprehensive repository of verified business identity, ownership, and corporate linkage data—directly into ChatGPT, Codex, Microsoft 365 Copilot, and Anthropic’s Claude in June 2026. The move marks one of the most significant steps yet to inject authoritative third-party data into generative AI tools, addressing the persistent problem of hallucinated business facts in enterprise workflows.
By weaving its data directly into the fabric of these large language models, Dun & Bradstreet aims to create a new trust layer for agentic workflows, where AI systems act on behalf of users to perform transactions, check compliance, or initiate supply chain actions. The integration promises to transform how procurement teams vet suppliers, how sales teams qualify leads, and how compliance officers monitor third-party risk—all within the familiar interfaces of Copilot, ChatGPT, and Claude.
The Commercial Graph: What It Contains and Why It Matters
The Dun & Bradstreet Commercial Graph is more than a static database. It is a living knowledge graph encompassing over 500 million business entities worldwide, including their legal names, doing-business-as (DBA) aliases, corporate family trees, beneficial ownership structures, and financial health indicators. Updated constantly through public records, government filings, and proprietary sources, it serves as a single source of truth for business identity.
In the context of generative AI, this graph acts as a grounding mechanism. When a user asks Copilot to “summarize the ownership structure of a potential supplier,” the model can query the Commercial Graph via a secure API, retrieve the verified data, and present it with confidence scores. This stands in stark contrast to the current default behavior of LLMs, which may piece together fragmented or outdated information from the web, often resulting in conflation of similarly named entities or outright fabrication.
The integration includes built-in feedback loops. If a model cannot find a match in the graph, it can flag an entity for verification rather than invent data, triggering a workflow for a Dun & Bradstreet analyst to enrich the record. This human-in-the-loop design will be critical for maintaining accuracy in high-stakes decisions such as anti-money laundering checks or credit assessments.
Inside the Integration: How It Works Across Platforms
Dun & Bradstreet’s approach leverages each platform’s native plugin or function-calling architecture. For OpenAI’s ChatGPT and Codex, the Commercial Graph will be accessible as a first-party API tool, allowing the model to call verified business data within a chat session or when generating code related to business logic. Developers using Codex can, for instance, write functions that automatically include D-U-N-S numbers, corporate hierarchies, or credit rankings into their applications without manually pulling from a separate database.
For Microsoft 365 Copilot, the integration runs deeper. Users drafting a request for proposal (RFP) in Word can command Copilot to “insert a company profile for Acme Corp including its parent entities and recent financial risk indicators,” and the graph will supply the data directly into the document. In Excel, financial analysts can call up a supplier’s payment history and risk metrics without leaving the spreadsheet, while in Teams, compliance officers can query a partner’s sanctions status in real time during a call.
Anthropic’s Claude will also tap into the graph, with an emphasis on safety and alignment. Claude’s constitutional AI framework means the model will be trained to cite sources from the Commercial Graph explicitly and to refuse to generate unverified business claims when the data is absent. This aligns with Claude’s design philosophy of being helpful, honest, and harmless.
A critical architectural feature is that the graph is read-only from the AI models’ perspective. The integrations are designed so that the LLMs cannot overwrite or corrupt the underlying data, only query it. This preserves data integrity and gives enterprises confidence that the information has not been tampered with by a hallucination or adversarial prompt.
The Next Data Trust Layer: Governance and Compliance
Dun & Bradstreet is positioning this as more than just a data feed; it’s a governance layer. The company’s proprietary Data Trust Framework assigns a confidence score to every piece of information, based on recency, source reliability, and corroboration. When a model retrieves data, it also pulls these trust metrics, enabling the AI to qualify its answers with phrases like “high confidence, last verified Q2 2026” or “low confidence, requires manual review.”
This touches on a major pain point for regulated industries. Financial services firms, pharmaceutical companies, and government contractors often cannot use AI for certain tasks because the model’s answers cannot be audited or traced back to a verified source. Dun & Bradstreet’s integration creates a clear lineage: every piece of business data cited by Copilot or ChatGPT can be traced to a specific record in the Commercial Graph and, by extension, to a government filing or public record.
Auditability is further strengthened by logging every query. For organizations bound by frameworks like SOX, GDPR, or the EU AI Act, this means they can produce a record of what data was used, when, and by which model, providing a defensible trail for regulators.
The integration also introduces what Dun & Bradstreet calls “policy-aware retrieval.” Companies can configure rules within the graph layer to prevent certain sensitive data from ever being shown to users of a particular AI assistant. For example, a company might restrict the display of beneficial ownership details to only certain job roles within Copilot, or block the export of credit-risk data to customer-facing chat interfaces.
Real-World Use Cases That Move the Needle
The most immediate wins will come in procurement and sales. A procurement manager drafting a contract can ask Copilot to “risk‑assess all third-tier suppliers in our European logistics chain,” and the system will traverse the Commercial Graph’s ownership links, pulling financial risk scores for each entity and summarizing them in a dashboard. This replaces a process that today can take weeks of manual research.
For sales teams, the integration turns ChatGPT into a powerful prospecting tool. Instead of searching LinkedIn or manually cross-referencing websites, a sales rep can prompt: “Find all subsidiaries of Big Tech Corp that have a high risk of late payment and are located in the Midwest,” and receive a verified list with D-U-N-S numbers and recent credit indicators. Because the data is authoritative, the rep can immediately act on it.
Mergers and acquisitions research will also benefit. Analysts can use Claude’s longer context window to ingest entire ownership structures, compare them against sanctions lists, and produce memos that cite actual Dun & Bradstreet records, drastically reducing the time—and error rate—of due diligence.
A subtler but equally important use case is in internal master data management. Large enterprises often suffer from duplicate or inconsistent supplier records across systems. By making the Dun & Bradstreet graph the single source of truth, AI agents in tools like Copilot or Codex can automatically reconcile entity records, merge duplicates, and flag out-of-date information, all with a human approval step.
Competition and Market Context
Dun & Bradstreet’s move puts it in direct competition with other business‑data providers like ZoomInfo and Moody’s Analytics, which are also racing to embed their data into AI tools. However, D&B’s vast database and its long-standing relationship with Microsoft give it a head start. Microsoft already uses D&B data in its own CRM and ERP platforms, and this new Copilot integration deepens that partnership.
At the same time, smaller startups are emerging with retrieval‑augmented generation (RAG) solutions specifically for business data. Dun & Bradstreet’s play preempts them by offering a turnkey, low-latency integration that requires no bespoke RAG pipeline for the user. The ability to simply turn on a toggle in the Copilot admin center or add a plugin in ChatGPT removes a major adoption hurdle.
Nevertheless, the success of such integrations will depend on pricing. Neither company has disclosed the cost model, but it is likely that organizations will need a Dun & Bradstreet subscription and possibly a per‑query fee for AI access. For large enterprises, the cost of bad data can outweigh the subscription, but for mid‑market firms, the price point will be critical to widespread adoption.
Challenges and Open Questions
No data integration is without risk. The primary concern is latency: if each query requires a live API call to the Commercial Graph, response times could slow, harming the fluidity of AI interactions. Dun & Bradstreet says it is caching frequently accessed records at the edge to maintain sub‑second latency, but this remains to be tested at scale.
Privacy is another issue. The Commercial Graph contains some information that is public, such as registered business addresses, but also derived scores and risk indicators that are proprietary. Models must be carefully prompted not to reveal the underlying logic or to inadvertently expose sensitive metrics in a chat that could be seen by unauthorized users.
Data freshness is a perennial headache. While the graph is updated continuously, there can be a lag between a change in government records and its reflection in the database. For use cases like verifying a company’s good standing, even a 24‑hour delay could be problematic. The integration includes a “freshness indicator” that warns users if a record is older than a set threshold, but enterprises will need to configure these thresholds to match their risk tolerance.
There is also the question of platform lock‑in. By building critical workflows around D&B’s data and these specific AI assistants, companies may find it difficult to switch providers later. Dun & Bradstreet and its platform partners will need to offer interoperability and open APIs to ease concerns.
Finally, the integration does not solve the fundamental problem of model reasoning. The LLMs can still mishandle the retrieved data—misinterpreting a parent‑subsidiary relationship or misapplying a risk rule. Human review remains essential for high‑stakes decisions, and Dun & Bradstreet emphasizes that the AI should assist, not replace, professional judgment.
What This Means for the Future of Agentic AI
The Dun & Bradstreet integration is a bellwether for the next phase of enterprise AI: moving from general‑purpose chat to trusted, action‑oriented agents. When an AI cannot only answer a question but also initiate a credit check, file a compliance report, or update a supplier record—all backed by verified data—the role of the knowledge worker changes fundamentally.
Companies will begin to build “agents” in Copilot Studio or Codex that can autonomously monitor supplier lists, flag risks, and even draft remediation plans. With a trust layer in place, these agents can operate with clear boundaries and auditable outputs, satisfying both internal audit and external regulators.
For Microsoft, this deepens the Copilot ecosystem into a platform for business data services, not just productivity. By embedding D&B’s data, Microsoft signals that it intends to make Copilot the hub for all enterprise‑grade AI tasks, from document creation to supply‑chain management.
OpenAI and Anthropic stand to gain as well, by showing that their models can be safely deployed in high‑compliance environments, a market segment that has been slow to adopt generative AI.
The integration also sets a template for other data providers. If the Dun & Bradstreet model works, we can expect similar embeds from companies like LexisNexis for legal data, IQVIA for clinical information, and Thomson Reuters for regulatory content—all within the same AI assistants.
Looking Ahead: The June 2026 Launch
The June 2026 rollout will be phased. Early access will go to existing large enterprise customers of Dun & Bradstreet and Microsoft, with general availability to follow. Dun & Bradstreet has promised detailed documentation, sandbox environments, and certification programs for developers and administrators to ensure proper governance from day one.
The company is also working on a feedback mechanism that allows users to report inaccuracies directly from within the AI interface, further closing the loop on data quality.
As generative AI leaves the novelty phase and enters the nerve center of business operations, the ability to trust its outputs becomes non‑negotiable. Dun & Bradstreet’s Commercial Graph integrations represent one of the most concrete steps yet toward a future where AI not only talks like a business analyst but also works from the same verified facts.