The House Veterans' Affairs Committee grilled Oracle executives over the Oracle VA EHRM contract increase, veteran data use and AI, a signal to every government contractor.
FedScoop reported the hearing, which took place Thursday, Nov. 19, 2026, before the House Veterans' Affairs Committee. The story matters far beyond the Department of Veterans Affairs. If you sell to the federal government, especially in a regulated industry handling health, defense or personal data, the questions lawmakers asked Oracle are the questions you should expect about your own contracts, your data practices and your AI ambitions.
This is Part 1 of a two-part analysis from Petronella Technology Group, Inc. Part 1 covers what happened, how the contract and the data dispute work, and what the evidence does and does not show. Part 2 will cover what regulated businesses and contractors should do about it.
Key Takeaways
- Oracle co-CEO Mike Sicilia and Executive Vice President Seema Verma testified before the House Veterans' Affairs Committee about a $17 billion increase to a $10 billion Department of Veterans Affairs electronic health record modernization contract, after a unanimous September subpoena compelled Sicilia's appearance.
- Lawmakers from both parties challenged Oracle on whether it wants veteran health data to train AI systems. Verma testified Oracle has "zero access" to veteran data; Rep. Nikki Budzinski said Oracle's sales team has pursued commercial data privacy terms that would "virtually allow you to own this data."
- Incoming committee chair Mark Takano called the bipartisan scrutiny "a really good signal" to all government contractors, saying everyone doing business with the government "has got to do it on the straight and narrow."
- Rep. Nancy Mace pressed Verma for an exact count of patient safety tickets older than 90 days; Verma estimated between 1,000 and 1,800 and said they are resolved in weeks if not days, an answer Mace rejected after the hearing.
- Rep. Abe Hamadeh said no further money should move until Oracle and the VA produce a Government Accountability Office-accepted cost estimate, a fixed deployment schedule, and a clear accounting of what has already been wasted.
What happened: the facts in detail
Two days of testimony, one subpoena
Oracle co-CEO Mike Sicilia and Executive Vice President Seema Verma testified Wednesday before the Senate Veterans' Affairs Committee, then appeared Thursday, Nov. 19, 2026, before the House Veterans' Affairs Committee. The House hearing ran two hours and featured bipartisan scrutiny. Sicilia appeared as a result of a unanimous September subpoena, according to FedScoop's report.
The contract increase under challenge
At the center of the hearing is a $17 billion increase to a $10 billion Department of Veterans Affairs electronic health record modernization contract. Chair Mike Bost, R-Ill., noted the expansion: "Any other contract with this size would have came through Congress, but now through both Republicans and Democrats, it has been expanded." He added, "You do understand it's over $2,000 per veteran. That's a lot of money."
Before the House panel, the executives reiterated the reasoning they had given the Senate: the contract ceiling increase was due to an expansion of scope and scale, COVID-19 delays, early customization and inadequate infrastructure. Verma also said 40% of the original contract's ceiling was spent by the time Oracle acquired Cerner in 2022.
Sicilia's defense
Sicilia told the representatives, "I'm not here to milk the taxpayers or the veterans. If there are other ways to reduce cost here, my main mission is to deliver the solution for veterans. I believe it has value for veterans, but if the government doesn't feel that way, then they should by no means continue to fund the program. But if we all feel that way, I'm willing to sit with any member of any committee. I'm willing to sit with the VA. I'm willing to sharpen every single pencil we can."
The data and AI accusations
The sharpest exchanges concerned veteran health data and artificial intelligence. Lawmakers accused Oracle of wanting veteran health data to train its AI systems. Verma's response was categorical: "We have zero access to the data of the veterans. We have testified multiple times. We have no access. We do not use that data for any purpose. It is only used by the VA to deliver care for veterans."
Rep. Nikki Budzinski, D-Ill., said Oracle's sales team "has been pursuing the inclusion of your commercial terms and conditions as it relates to data privacy, which would virtually allow you to own this data." Takano said the VA told Congress it had a "long and feisty" discussion with Oracle about the use of veteran data, which the executives said they were not aware of.
Takano went further, saying, "It's clear that Trump's allies were well aware of the value of this data 10 years ago, which has only become more valuable with the growth of large language models and generative AI. This deal set up what could become the most significant commercial exploitation of veterans' data ever seen."
Patient safety tickets and service levels
Rep. Nancy Mace, R-S.C., questioned why Verma could not provide the exact number of patient safety tickets older than 90 days. Verma estimated it is between 1,000 and 1,800 and said they are resolved in weeks if not days. Mace was not satisfied, telling FedScoop after the hearing that Verma should have shown up with answers: "Yeah, as a software company, you should be able to talk about software numbers."
On performance, Sicilia said the contract went from eight service level agreements to 22 and that "We have largely met those service level agreements. There is one exception in May of 2025 where we did not meet the system uptime agreement, and we paid the penalty." He also said the next version of the software is included at no additional cost, and that Oracle would "love the VA to roll out the most modern technology in the world," including the company's clinical AI agent. Verma said rollout progress is not perfect but is still significant.
Lawmakers who want the money stopped
Rep. Abe Hamadeh, R-Ariz., called the speed of the rollout "embarrassing" and said he is "appalled." His demand: "No further money should move until Oracle and the VA produce a [Government Accountability Office]-accepted cost estimate, a fixed deployment schedule, and a clear accounting of what has already been wasted."
The bipartisan "signal"
Mark Takano, the assumed incoming House Veterans' Affairs Committee chair, told FedScoop the hearing was "refreshing" and a "really good signal" to other government contractors. "There's a new ethos in Washington, and it's potentially bipartisan," he said. "It's not just Democrats getting even, it's not us seeking vengeance against Donald Trump, it's about all the contractors. Everyone who does business with this government has got to do it on the straight and narrow."
How the Oracle VA EHRM dispute works, and why it matters technically
Strip away the politics and the Oracle VA EHRM fight is about three things any regulated business recognizes: scope growth on a large technology program, control of sensitive data, and the arrival of AI into a system of record.
First, scope. The executives attributed the ceiling increase to expanded scope and scale, pandemic delays, early customization and inadequate infrastructure. Anyone who has run a large electronic health record or enterprise system migration will recognize that list. Customization and infrastructure gaps are where fixed expectations become open-ended costs. The relevant question for a buyer is not only what the software costs, but who absorbs the cost of the environment not being ready, and whether the contract structure makes that visible before the money is spent. The hearing shows Congress now asking that question in public, with a per-beneficiary figure, over $2,000 per veteran by Bost's count, attached to it.
Second, data rights. Verma's testimony and Budzinski's accusation describe two different contract worlds. In Oracle's commercial world, data privacy terms and conditions can, in Budzinski's words, "virtually allow you to own this data." In the government world, the expectation is the opposite: the customer, here the VA, holds the data, and the vendor processes it only to deliver the service. Verma testified Oracle has zero access to veteran data and uses none of it for any purpose. The unresolved dispute is whether Oracle's commercial contracting habits were being carried into a government deal, and what the VA's own "long and feisty" discussion with Oracle about data use actually covered. For any business negotiating with a large software vendor, this is the clause to read twice: who can use the data, for what, and who owns derived outputs such as model training artifacts.
Third, AI. Takano's point about large language models and generative AI is the technical heart of the story. Health records are among the most valuable training corpora that exist, and their value has grown as generative AI has matured. Sicilia, for his part, volunteered that Oracle wants the VA to adopt its clinical AI agent and that the next software version comes at no additional cost. Those two facts sit in tension: a vendor that says it has no access to the data, and a vendor that wants its AI embedded in the system that holds the data. Both can be true if the AI runs inside the customer's boundary on the customer's data, with the vendor's models and the customer's records kept separate. Whether that is how the Oracle arrangement works is exactly what the hearing did not resolve. Organizations wrestling with that same boundary question, keeping sensitive records inside a controlled environment while still using modern AI, can read our guide to HIPAA-Compliant Private LLMs: 5 Architectures and our analysis of Secure Enterprise RAG: Vector DB and LLMOps.
Finally, accountability mechanics. The contract moved from eight service level agreements to 22, and Sicilia said Oracle has largely met them, with one exception in May of 2025 on system uptime for which Oracle paid a penalty. Service level agreements are how performance becomes enforceable in software contracts. The Mace exchange over patient safety tickets older than 90 days shows the other side: if a vendor cannot produce a precise operational number on demand, lawmakers and customers will treat the absence itself as a finding. Hamadeh's demand for a Government Accountability Office-accepted cost estimate and a fixed deployment schedule is, in effect, a demand to convert a flexible program back into a measurable one.
What the evidence does and does not show
It is worth being precise about what the public record establishes and where it stops.
What is established. The House Veterans' Affairs Committee held a two-hour hearing on Nov. 19, 2026, with Sicilia and Verma testifying under bipartisan questioning, Sicilia appearing after a unanimous September subpoena. The contract figures, the $17 billion increase to a $10 billion contract and the over $2,000 per veteran figure, come from the hearing as reported by FedScoop. Verma's testimony that 40% of the original ceiling was spent by the time Oracle acquired Cerner in 2022 is on the record, as are the stated causes: scope and scale expansion, COVID-19 delays, early customization and inadequate infrastructure. Sicilia's statements on service level agreements, the May 2025 uptime exception and the paid penalty are also on the record.
What is disputed. The data question is unresolved. Verma testified Oracle has zero access to veteran data and uses none of it. Budzinski asserted Oracle's sales team has pursued commercial data privacy terms that would virtually allow Oracle to own the data. Takano said the VA described a "long and feisty" discussion with Oracle about veteran data use, which the executives said they were not aware of. These are competing accounts from a single hearing. The sources do not include the contract language itself, so no reader of this coverage can independently verify which characterization of the data terms is accurate.
What is unknown. Several things the hearing surfaced were not answered there. The exact number of patient safety tickets older than 90 days was not provided; Verma gave an estimated range of between 1,000 and 1,800. What the VA and Oracle actually discussed regarding veteran data use is not public. Whether Congress will impose Hamadeh's conditions, a GAO-accepted cost estimate, a fixed deployment schedule and an accounting of wasted funds, is not known. The future of funding for the program itself remains an open question; Sicilia himself said that if the government does not feel the program has value, "they should by no means continue to fund the program."
What the coverage cannot tell you. FedScoop's account captures the hearing, the written reasoning the executives offered, and post-hearing comments from Mace and Takano. It does not include the underlying contract documents, the VA's internal position beyond what lawmakers relayed, or any independent audit of the program's performance. Takano's characterization of what "this deal set up" is his assessment, not a documented finding. Readers should hold that distinction firmly: the hearing established that lawmakers believe the data terms are dangerous; it did not establish what the data terms are.
That gap, between public accusations and contract text, is precisely why this story matters to contractors, and it is where Part 2 of this analysis picks up.
The Compliance and Risk Implications for Regulated Organizations
The Oracle VA EHRM hearing is not just a procurement story. It is a preview of how Congress, regulators, and the public now treat any organization that holds sensitive data on behalf of the government or on behalf of patients. Three themes from the hearing map directly onto obligations that regulated businesses already carry.
First, data ownership and data use terms are now front-page issues. Lawmakers accused Oracle of pursuing commercial terms that would, in Rep. Nikki Budzinski's words, "virtually allow you to own this data," and of wanting veteran health data to train AI systems. Oracle's Seema Verma testified that Oracle has "zero access to the data of the veterans" and does not use it for any purpose. Whatever the truth, the lesson is clear: the contract language governing who can access data, for what purpose, and whether it can feed AI models, will be read aloud in a public hearing. Any business that signs vendor agreements without negotiating data use restrictions is accepting the same risk Oracle faced on the dais.
Second, accountability now extends to operational metrics you must be able to produce on demand. Rep. Nancy Mace pressed Verma on why she could not give the exact number of patient safety tickets older than 90 days; Verma estimated between 1,000 and 1,800. Mace's post-hearing comment, that a software company "should be able to talk about software numbers," is the same expectation regulators bring to a HIPAA risk analysis, a CMMC assessment, or a PCI DSS review. If you cannot produce your own numbers with evidence behind them, the gap between your estimate and your documentation becomes the story.
Third, service level commitments are becoming enforcement instruments. Sicilia testified that the contract grew from eight service level agreements to 22, that Oracle has "largely met" them, and that there was one exception in May of 2025 where Oracle missed the system uptime agreement and paid the penalty. For regulated organizations, uptime and incident response commitments increasingly appear in both government contracts and business associate agreements. A commitment you cannot measure, or cannot meet, is a liability with a timestamp.
These themes connect to frameworks many readers already live under. Defense contractors handling controlled unclassified information operate under NIST SP 800-171 and, where DFARS 252.204-7021 appears in a contract, CMMC certification requirements. Healthcare organizations operate under the HIPAA Security Rule at 45 CFR Part 164, which requires a risk analysis and administrative, physical, and technical safeguards. PCI DSS v4.0.1 covers 12 control areas for cardholder data environments. The common thread with the Oracle VA EHRM dispute is that all of these frameworks demand documented control over who touches sensitive data and proof that promised safeguards actually operate.
What This Means for Regulated Industries
Defense Contractors
Takano's framing, that "everyone who does business with this government has got to do it on the straight and narrow," is a direct signal to the defense industrial base. Contractors should treat the Oracle VA EHRM hearing as evidence that congressional scrutiny of contractor security and data handling is bipartisan and intensifying. If your NIST SP 800-171 self-assessment score in SPRS does not match reality, or your system security plan describes controls that exist only on paper, you are carrying the same exposure Oracle executives faced, minus the cameras.
Healthcare
The accusation at the center of the hearing, that a vendor might use patient data to train AI, is the exact fear driving healthcare privacy enforcement today. Business associates accounted for about 66% of breached healthcare records in 2024, and Change Healthcare's breach ultimately reached 192.7 million individuals in the final HHS filing of July 2025. Healthcare organizations should review every vendor agreement for data use clauses, confirm that PHI is used only for care delivery or contracted services, and ensure AI vendors contractually commit to no model training on patient data. IBM's 2026 figure puts the average healthcare breach at $6.64 million, and HHS OCR enforcement actions in 2025 and 2026, including a $3 million Solara settlement and a $1.165 million April 2026 ransomware sweep, show regulators are actively penalizing gaps. Our guide to HIPAA violation penalties and enforcement covers the current penalty landscape in detail.
Legal
Law firms and legal technology vendors sit in a position similar to Oracle's in this story: trusted custodians of highly sensitive client data with growing pressure to adopt AI. The hearing's central question, whether the custodian can prove it has "zero access" to the data it stores, is the question sophisticated clients now ask counsel. Firms deploying AI tools should be able to demonstrate, contractually and technically, that client matter data does not train third-party models, that access is logged, and that the firm can produce its own metrics on data handling when a client or regulator asks. Self-hosted and private AI architectures are often the cleanest answer; our analysis of HIPAA-compliant private LLM architectures applies equally to legal confidentiality obligations.
Financial Services
Financial institutions face the same convergence: large vendor contracts, valuable consumer data, and AI ambitions under regulatory watch. The Oracle VA EHRM hearing shows that oversight bodies will question not just whether data is protected, but whether contract terms quietly expand vendor rights to it. Financial firms should apply the same discipline to their own vendor management: inventory which processors hold customer data, review the data use and ownership language in each agreement, and require vendors to report measurable service metrics. Regulators in this sector have long expected vendor accountability; this hearing shows the expectation going mainstream.
An Action Plan for This Month
Security and IT leaders can act on the lessons of this hearing without waiting for new regulations. Here is a practical sequence:
- Pull every vendor contract that touches sensitive data. Identify agreements where data use, data ownership, or AI training rights are vague or missing. These are your highest-priority renegotiations.
- Add or confirm a no-training clause. For any vendor with access to PHI, CUI, or customer data, require explicit contractual language that your data will not be used to train AI models, and that the vendor holds no rights to it beyond delivering the service.
- Inventory your AI exposure. Document every AI tool in use, including shadow AI adopted by employees without approval. IBM's 2025 research found 20% of breaches involved shadow AI, meaning unmanaged tools are already a breach vector.
- Test whether you can produce your own numbers. Pick three operational metrics a regulator or customer would ask about, such as open security tickets, incident response times, or access review completion, and verify you can produce them with evidence within one business day.
- Review your service level commitments. List every SLA you have promised to customers or the government, including uptime and incident response, and confirm you can measure and meet each one. Know your penalty exposure.
- Reconcile your self-assessment against reality. If you have posted a NIST SP 800-171 score in SPRS or signed a HIPAA risk analysis, walk a sample of claimed controls and confirm each one operates as documented.
- Tighten access logging on sensitive data stores. The claim that a vendor has "zero access" is only defensible if access is technically enforced and logged. Verify that audit logging on your most sensitive systems would actually prove who touched what.
- Establish an incident reporting readiness check. If you are a defense contractor, confirm you can meet the DFARS 252.204-7012 requirement to report a cyber incident affecting covered defense information within 72 hours, including having the required DoD-approved medium assurance certificate obtained in advance.
- Brief leadership on data governance risk. The Oracle VA EHRM story became a congressional hearing partly because data terms were negotiated without sustained executive attention. Put data use rights on the agenda of your next board or leadership meeting.
- Get an outside read on your gaps. A short independent assessment of your data governance and AI exposure, against the frameworks that apply to you, will find what internal review misses.
How Petronella Technology Group, Inc. Helps
Petronella Technology Group, Inc. works with regulated businesses and defense contractors that face exactly the pressures this hearing exposed: sensitive data, AI ambitions, and auditors or oversight bodies asking for proof.
On the compliance side, our compliance services cover CMMC readiness, HIPAA, SOC 2, and related frameworks. A typical CMMC Level 2 readiness engagement runs 12 to 14 weeks for a mid-size contractor, moving through discovery, gap analysis, a remediation sprint, and C3PAO readiness handoff, producing a system security plan covering all 110 NIST SP 800-171 controls, control-family policies, a live POA&M, and a mock assessment. Every practitioner on our compliance team holds the CMMC-RP designation, and we are listed on the Cyber AB Marketplace as Registered Provider Organization #1449. We prepare contractors for assessment; only an authorized C3PAO can certify.
For organizations that need ongoing security leadership without a $325K to $450K loaded annual CISO hire, our vCISO service provides a named security officer of record for HIPAA, CMMC, SOC 2, and PCI responsibilities, with a written current-state assessment within 30 days and a first quarterly board memo at day 75 to 90.
On the operational side, our 24/7 managed detection and response service combines AI-powered analytics with human-led investigation, running on our own private AI cluster so that defense industrial base and healthcare clients never send CUI or PHI to a public-cloud SOC. The AI never closes a ticket on its own and every action is logged for CMMC and HIPAA audit. Our cybersecurity services layer a 39-plus layer security stack, including MFA, FIPS-validated encryption, and audit logging, over People, Process, and Technology.
For organizations that want AI capability without the data exposure at the heart of the Oracle VA EHRM dispute, we design, build, and operate private AI clusters that keep data inside your boundary, mapped to CMMC, HIPAA, and DFARS 252.204-7012 requirements. Our enterprise private AI deployment guide walks through the architecture decisions, and our work on secure enterprise RAG and vector database governance addresses the retrieval-layer controls that determine whether an AI system can leak what it should not.
Related reading
- Oracle Health Data Breach Tally Climbs to Nearly 20 Million
- Oracle Data Breach: Russian Hackers Strike POS Systems
- Maze Ransomware Leaks Patient PTSD Records from Law Firms
Sources
- A ‘signal’ to government contractors: House panel grills Oracle execs on VA EHRM, AI, data
- MPEs gain momentum for sharing information with allied partners
- CISA director pick already drawing concern from Hill Dems over Palantir connection
- Trump picks Palantir’s LoSapio to head CISA
- Cross-Sector Cybersecurity Performance Goals
- SP 800-61 Rev. 3, Incident Response Recommendations and Considerations for Cybersecurity R
Frequently Asked Questions
Why did the House grill Oracle executives about the VA EHRM contract?
The House Veterans' Affairs Committee questioned Oracle co-CEO Mike Sicilia and Executive Vice President Seema Verma about the $17 billion increase to a $10 billion Department of Veterans Affairs electronic health record modernization contract. Chair Mike Bost noted the expansion amounted to over $2,000 per veteran, and Sicilia appeared as a result of a unanimous September subpoena. The executives attributed the increase to expanded scope and scale, COVID-19 delays, early customization, and inadequate infrastructure.
Does Oracle use veteran health data to train its AI systems?
Oracle says no. Verma testified that Oracle has "zero access to the data of the veterans" and that the data is used only by the VA to deliver care. Lawmakers, including Rep. Budzinski and Rep. Takano, questioned whether Oracle's commercial contract terms would effectively allow it to own the data, and Takano warned the deal could become "the most significant commercial exploitation of veterans' data ever seen." The dispute over contract language, rather than a confirmed data use, is what the hearing surfaced.
What does the Oracle VA EHRM hearing mean for other government contractors?
Takano called the bipartisan scrutiny "a really good signal" to other contractors, saying everyone doing business with the government "has got to do it on the straight and narrow." For contractors, the practical meaning is that data handling practices, security posture, and contract terms are all fair game for oversight, and misrepresenting any of them carries fraud risk under initiatives like the DOJ Civil Cyber-Fraud Initiative.
How can a regulated business prevent vendors from using its data for AI training?
Start with contract language: require explicit terms that prohibit model training on your data and limit vendor rights to delivering the service. Then verify technically: enforce and log access controls so you can prove who touched the data, and prefer architectures where sensitive data stays within your own boundary, such as private AI deployments on infrastructure you control. Contract terms and technical enforcement together are what make a "zero access" claim defensible.
What should healthcare organizations learn from this story?
The hearing shows that data use terms in vendor agreements are now a public accountability issue, not a legal footnote. Healthcare organizations should review business associate agreements for AI training restrictions, confirm PHI is used only for permitted purposes, and be able to produce operational metrics on demand. With business associates involved in about 66% of breached healthcare records in 2024 and the average healthcare breach costing $6.64 million per IBM's 2026 research, vendor data governance is a direct financial risk.
Is the VA EHRM rollout stopping?
Not based on what is known. Some lawmakers, including Rep. Abe Hamadeh, said no further money should move until Oracle and the VA produce a GAO-accepted cost estimate, a fixed deployment schedule, and a clear accounting of what has been spent. Sicilia testified that Oracle has largely met the contract's 22 service level agreements, with one uptime exception in May 2025 for which it paid a penalty, and said the next software version is included at no additional cost. Whether funding continues remains a decision for Congress and the VA.
If the questions this hearing raises about your own data governance, AI exposure, or compliance posture feel familiar, we can help you get concrete answers. Call Penny, our AI assistant, at 919-348-4912 to book a consultation, or reach Petronella Technology Group, Inc. through our contact form.
Free online course from the Petronella Technology Group, Inc. Training Academy
Free crash course: 5 practical steps to protect your data and meet HIPAA, CMMC, PCI-DSS, and FTC Safeguards compliance standards.
Price: Free
Start the free course Cybercrime and Identity Theft Defense Mastery ($997)
Free, practical, and specific to regulated environments. We will email it to you.
No spam. Unsubscribe anytime.