Blog
Who is accountable when AI is wrong? What the OpenAI–Anthropic–Department of Defense standoff means for enterprise systems
I have spent the last several weeks head down in CMMC. And with the latest developments at the DoD, it begs the question, who is accountable?
Artificial intelligence has moved decisively out of experimentation and into the core of enterprise platforms. ERP systems, once slow‑moving back‑office tools, are now becoming AI‑assisted operational engines. Recent events involving OpenAI, Anthropic, and the U.S. Department of Defense make one thing clear: AI is no longer just a productivity feature. It is now a governance, compliance, and national‑interest concern.
For ERP leaders, CIOs, and executives, this moment should trigger a reset in how AI inside business systems is evaluated, deployed, and governed.
What Just Happened, and Why It Matters to ERP
Over the weekend, the Department of Defense finalized an agreement with OpenAI to deploy advanced AI models within classified defense networks. This followed a breakdown in negotiations with Anthropic.
Anthropic refused to loosen contractual guardrails that would allow its models to be used for mass domestic surveillance or fully autonomous weapons decisions. As a result, Anthropic was designated a supply‑chain risk, and federal agencies were ordered to phase out its technology by 5:01 PM on Friday.
OpenAI, by contrast, accepted deployment under strict but government‑approved safeguards. These included human oversight for use‑of‑force decisions, prohibitions on domestic mass surveillance, and cloud‑only, auditable deployment architectures.
These details matter, not because most ERP systems will ever run on classified military networks, but because this is now the benchmark for acceptable enterprise AI behavior.
ERP Is Becoming a Decision System, Not Just a System of Record
Modern ERP platforms, especially cloud ERPs like Dynamics 365, are rapidly embedding AI copilots, agentic workflows, and predictive and prescriptive recommendations. This shifts ERP from record‑keeping to decision‑shaping. Today’s ERP AI can recommend actions, automate workflows, interpret unstructured data, and influence financial, operational, and supply‑chain outcomes.
Once AI starts influencing outcomes rather than simply summarizing data, the conversation shifts from features to accountability. Accountability changes EVERYTHING.
What the Department of Defense is effectively saying is this: if AI influences decisions, governance cannot be optional. Guardrails must be explicit. Humans must remain accountable. Usage must be auditable.
That same expectation is coming to commercial ERP environments, especially in regulated industries, government contracting, and complex manufacturing operations.
The Compliance Signal: Guardrails Are Not Optional
The Department of Defense’s treatment of Anthropic sends a strong message to the commercial market. AI governance cannot be abstract or aspirational. It must be contractual, enforceable, and auditable.
For ERP environments, this has several implications.
1. AI Must Be Governed Like a User
AI agents will require explicit role definitions, least‑privilege access, approval workflows, and audit trails.
2. “Human in the Loop” Is No Longer Marketing Language
The OpenAI agreement with the Department of Defense codifies human responsibility for high‑impact decisions. In ERP terms, this means AI may prepare, recommend, or summarize, but humans must approve, execute, and own outcomes.
This distinction will increasingly matter in audits, litigation, and regulatory reviews.
3. Cloud Architecture and Data Residency Matter More Than Ever
The Department of Defense required controlled, cloud‑based deployments with vendor‑managed safety stacks. For ERP buyers, particularly those handling government contracts, controlled unclassified information, or regulated data, this reinforces the importance of FedRAMP alignment, clear data‑flow boundaries, and vendor transparency around model hosting and training.
Why This Should Concern Commercial ERP Buyers
Many organizations assume that government‑grade rules do not apply to them. History suggests otherwise. Security, privacy, and compliance standards almost always appear first in defense and intelligence, then move into regulated industries, and eventually become baseline expectations across the broader market.
We have seen this pattern with Zero Trust security, SOX controls, and data residency regulations. AI governance inside ERP is following the same trajectory.
A Practical Framework for ERP Leaders
Rather than reacting emotionally to headlines, ERP decision‑makers should ask four grounded questions:
- Where is AI already influencing decisions in our ERP?
- What data does it access, and what data must be restricted?
- Who is accountable when AI is wrong?
- Could we explain our AI controls to an auditor, regulator, or customer? Is it like ITGC’s?
- If those answers are unclear, the issue is not the AI model. It is the ERP foundation beneath it.
The Era of Casual AI in ERP Is Ending
The OpenAI, Anthropic, and Department of Defense conflict is not about politics or ideology. It is about control, accountability, and trust.
ERP systems sit at the intersection of finance, operations, supply chain, and compliance. As AI becomes embedded in these platforms, expectations will rise, whether organizations are ready or not. The companies that struggle will not be the ones that avoided AI. They will be the ones that adopted it casually, without governance, ownership, or clarity.
The winners will not be the fastest adopters. They will be the most deliberate ones. AI belongs in ERP. It delivers real value. But the era of casual AI inside core business systems is ending.
ERP leaders should treat this moment not as a warning, but as an opportunity to get it right.
if you are interested in learning how and when to deploy AI solutions that work for you.
