Latest / Encrypted Ambition: Where Ambition Meets Encryption - Inside The Minds Of Disruptors. / DeepSeek and the FY2026 NDAA

Transcript

DeepSeek and the FY2026 NDAA · Encrypted Ambition: Where Ambition Meets Encryption - Inside The Minds Of Disruptors.

• 0:14
  Host: Today we dive into the new DoD AI ban that forces contractors to rethink every model they use.
• 0:21
  Analyst: The law, Section 1532 of the FY2026 NDAA, prohibits any contractor from using AI developed by DeepSeek or High Flyer during DoD contract performance.
• 0:33
  Host: So if a company runs a DeepSeek model on its own servers, does the ban still apply?
• 0:38
  Analyst: Yes. The statute never mentions deployment mode. Whether the model is in the cloud or on-premises, it still counts as ‘use’.
• 0:46
  Host: That's surprising because many think self-hosting might be a loophole.
• 0:50
  Analyst: The law attaches to the developer, not the hosting arrangement. It follows the model wherever it runs.
• 0:56
  Host: So a DeepSeek API call or a GPU cluster in a private data center is covered.
• 1:01
  Analyst: Correct. The text says nothing about local-model exemption. The definition of ‘covered artificial intelligence’ runs on developer lineage.
• 1:10
  Host: What about fine-tuning a DeepSeek model? Does that change anything?
• 1:14
  Analyst: Fine-tuning still ties back to the base model. The statute includes successor AI developed by DeepSeek or High Flyer.
• 1:22
  Host: So any modification doesn't help avoid the prohibition.
• 1:25
  Analyst: Exactly. The prohibition applies to any use in contract performance, regardless of how the model is accessed.
• 1:33
  Host: What about other Chinese-origin models like Qwen or GLM? Are they covered?
• 1:38
  Analyst: The statute names only DeepSeek and High Flyer. Qwen, GLM, Kimi, MiniMax are not listed.
• 1:44
  Host: But I've heard Alibaba and Baidu were added to a DoD list. Does that change things?
• 1:50
  Analyst: They joined the Section 1260H list, which could trigger a broader prohibition if the Secretary issues guidance.
• 1:57
  Host: Has any guidance been issued yet?
• 1:59
  Analyst: No public confirmation exists. The wider category remains a watch item, not a live prohibition.
• 2:06
  Host: So for now, only DeepSeek and High Flyer are strictly prohibited.
• 2:10
  Analyst: Yes, but be aware that pending legislation could add more names, and the current list is contested.
• 2:16
  Host: Let's talk about compliance frameworks. Does CMMC or NIST 800-171 mention these AI restrictions?
• 2:23
  Analyst: Neither does. CMMC maps to 110 controls of NIST 800-171, but there is no AI-origin rule.
• 2:33
  Host: So the statutory ban is separate from CMMC compliance.
• 2:36
  Analyst: Correct. The ban applies to the use of covered AI, while CMMC requires adequate security controls for every system.
• 2:44
  Host: What about DFARS 252-7012? Does it involve AI models?
• 2:50
  Analyst: DFARS 252-7012 mandates that covered contractor systems meet the same 110 controls, and cloud providers must meet FedRAMP Moderate or equivalent.
• 3:03
  Host: So removing DeepSeek doesn't satisfy DFARS?
• 3:06
  Analyst: No. DFARS concerns control implementation, not model lineage. You still need to address the statutory prohibition separately.
• 3:16
  Host: Understood. Let's discuss the practical steps a contractor should take.
• 3:20
  Analyst: First, inventory every AI model in use, both commercial and self-hosted, and trace each against the covered AI definition.
• 3:29
  Host: How do you trace a model to the definition?
• 3:31
  Analyst: Check the developer lineage. If the model originates from DeepSeek or High Flyer, or an entity with at least 20 percent stake in High Flyer, it is covered.
• 3:41
  Host: What about models that are not named? Do they automatically comply?
• 3:45
  Analyst: Not automatically. They still need to be secured under the 110 controls. You must verify they don’t send CUI to external systems.
• 3:54
  Host: So the next step is mapping the environment to the 110 controls?
• 3:58
  Analyst: Exactly. Use the seven-stage method: define the data boundary, prototype with your data, size hardware, isolate the cluster, deploy the model, layer access controls, and validate.
• 4:10
  Host: Could you give an example of hardware sizing?
• 4:13
  Analyst: A 7B parameter model runs on a single NVIDIA A100 or H100 GPU. Larger models may need two to four GPUs.
• 4:23
  Host: What about open-weight models like Gemma 4? Are they suitable for DoD work?
• 4:28
  Analyst: Gemma 4 is released under Apache 2.0, and it is not named in any U.S. restriction. It can be a compliant replacement.
• 4:36
  Host: Does the license matter for procurement?
• 4:38
  Analyst: Yes. Apache 2.0 is commercially permissive, so your procurement team can handle it without additional legal hurdles.
• 4:45
  Host: What about compliance documentation?
• 4:47
  Analyst: Petronella Technology Group provides a compliance platform called ComplianceArmor®. It automates SSP authoring, POA&M tracking, and evidence repository organization.
• 5:00
  Host: So you can integrate the inventory and removal record into that system?
• 5:04
  Analyst: Exactly. That documentation feeds into your system security plan and ensures you have evidence before any audit.
• 5:12
  Host: What about the waiver mechanism? Can we get a waiver for using DeepSeek?
• 5:16
  Analyst: Section 1532(b) offers waivers only for scientific research, national security, counterterrorism, or mission-critical functions, with risk mitigation steps.
• 5:28
  Host: So ordinary contract analytics wouldn't qualify?
• 5:31
  Analyst: Correct. The categories are narrow, and the waiver is an affirmative government decision, not an internal exception.
• 5:38
  Host: If we can't get a waiver, what is the next best action?
• 5:41
  Analyst: Remove the model from any DoD contract workflow and replace it with an unnamed, compliant model like Gemma 4 or Llama 3.1.
• 5:50
  Host: Does that satisfy DFARS 252-7012?
• 5:55
  Analyst: Yes, provided you implement the 110 controls for the new model environment. DFARS is about security, not lineage.
• 6:03
  Host: What about the intelligence community? Does Section 6604 affect us?
• 6:08
  Analyst: Section 6604 prohibits DeepSeek on intelligence community systems, and it directs the Director of National Intelligence to develop standards.
• 6:18
  Host: So if we support an IC element, the ban applies even if the broader Section 1532 doesn't?
• 6:24
  Analyst: Correct. Section 6604 is application-focused and reaches IC contractors directly.
• 6:30
  Host: So we need to check whether we work with any IC element.
• 6:33
  Analyst: Yes. If you do, you must remove DeepSeek from those workflows and replace it with a compliant model.
• 6:40
  Host: Let's talk about the timeline. When do we need to act?
• 6:43
  Analyst: Section 1532(a)(1) required the Department to remove covered AI from its systems within 30 days of enactment. The same pace applies downstream.
• 6:54
  Host: So we have a 30-day window to audit our systems?
• 6:57
  Analyst: Yes, but the audit expectation is already explicit in practitioner guidance. You should start immediately.
• 7:04
  Host: What about the pending FY2027 bill? Should we wait for it?
• 7:08
  Analyst: The bill could add names like Baidu, Zhipu AI, Moonshot AI, 01.AI, MiniMax, Alibaba, and Tencent. Until it becomes law, treat it as a watch item.
• 7:19
  Host: So we should monitor guidance and stay ready to adjust?
• 7:22
  Analyst: Exactly. Set a review cadence for any new guidance or legislation, and keep your compliance documentation up to date.
• 7:30
  Host: What if we accidentally use a covered model in a non-DoD contract?
• 7:34
  Analyst: The prohibition is limited to DoD contract performance. Non-DoD contracts are outside the statutory scope.
• 7:41
  Host: But could that still trigger a flow-down clause or other contractual obligation?
• 7:46
  Analyst: It depends on the specific DFARS clause. The statute didn't create a new clause, so you need to consult counsel.
• 7:53
  Host: So it's safest to audit all AI use regardless of contract type.
• 7:57
  Analyst: Yes, that proactive approach reduces risk and ensures compliance across all operations.
• 8:03
  Host: How do we document the removal of a covered model?
• 8:06
  Analyst: Record the model version, the date of removal, the reason, and the replacement model details in your system security plan.
• 8:14
  Host: And we should keep evidence of the waiver if we ever apply for one?
• 8:17
  Analyst: Yes, maintain signed attestations, risk mitigation plans, and any official waiver documents.
• 8:24
  Host: What are the key takeaways for a contractor right now?
• 8:27
  Analyst: Audit all AI, remove covered models from DoD workflows, replace with compliant open-weight models, map to the 110 controls, and stay alert to new guidance.
• 8:38
  Host: What are the deeper implications if a contractor ignores Section 1532?
• 8:44
  Analyst: Ignoring it could trigger enforcement actions, penalties, and loss of DoD contracts.
• 8:49
  Host: Could those penalties be monetary or purely contractual?
• 8:53
  Analyst: Both; the law can impose fines and force contract termination.
• 8:57
  Host: Is there a risk of a broader security audit?
• 8:59
  Analyst: Yes, auditors may scrutinize all AI systems tied to DoD work.
• 9:04
  Host: How does this affect non-DoD data handling?
• 9:06
  Analyst: Non-DoD data can still be processed, but the model must be compliant.
• 9:11
  Host: What about the hardware cost for a compliant model?
• 9:14
  Analyst: A 7B parameter model fits on a single NVIDIA A100, but larger ones need multiple GPUs.
• 9:21
  Host: Do we need to size the cluster before choosing a model?
• 9:24
  Analyst: Yes, a scoping prototype on your data informs hardware sizing.
• 9:29
  Host: What if we have a mix of models, some covered, some not?
• 9:32
  Analyst: Separate them by workflow; remove covered ones from DoD paths.
• 9:36
  Host: Can we still use a covered model for training only?
• 9:39
  Analyst: Section 1532(b) allows case-by-case waivers for research or training.
• 9:45
  Host: How do we apply for such a waiver?
• 9:47
  Analyst: Submit a formal request with risk mitigation steps; approval is government-issued.
• 9:53
  Host: If we get denied, what then?
• 9:55
  Analyst: You must replace the model before continuing DoD work.
• 9:58
  Host: What are the common mistakes we should avoid?
• 10:01
  Analyst: Assuming local hosting is exempt; assuming fine-tuning removes coverage.
• 10:05
  Host: Is fine-tuning a new model under the law?
• 10:08
  Analyst: No, the base model remains covered regardless of modifications.
• 10:12
  Host: How do we document that we have removed a covered model?
• 10:15
  Analyst: Log the version, removal date, reason, and replacement details in the system security plan.
• 10:22
  Host: Do we need to update the system security plan for every change?
• 10:25
  Analyst: Yes, keep the plan current to reflect the latest AI environment.
• 10:29
  Host: What about the 30-day window for removal?
• 10:31
  Analyst: The statute gave the Department 30 days; contractors should act promptly.
• 10:36
  Host: Should we monitor for new guidance from the Secretary?
• 10:39
  Analyst: Absolutely; a review cadence every 60 days keeps you aligned with Section 1532(a)(2).
• 10:46
  Host: How does CMMC Level 2 interact with this requirement?
• 10:50
  Analyst: CMMC Level 2 maps to NIST 800-171 controls but does not lift the statutory ban.
• 10:58
  Host: So we still need to remove covered models even with a CMMC certificate?
• 11:02
  Analyst: Correct; compliance frameworks and statutes operate independently.
• 11:06
  Host: What if we use a model from an unnamed Chinese-origin company?
• 11:10
  Analyst: Check the 1260H list and pending Section 1651; these may be added later.
• 11:16
  Host: Does the 1260H list immediately impose a ban?
• 11:20
  Analyst: Only if the Secretary issues guidance under Section 1532(a)(2); currently, it's a watch item.
• 11:27
  Host: Could a vendor's 20 percent stake trigger coverage?
• 11:30
  Analyst: Yes, any entity with at least 20 percent stake in High Flyer is covered.
• 11:35
  Host: What about the 24/7 AI-plus-human hybrid threat analysis?
• 11:39
  Analyst: That approach keeps AI in a controlled environment but still requires model compliance.
• 11:45
  Host: How do we ensure the AI stays within a controlled enclave?
• 11:48
  Analyst: Use a segmented VLAN or air-gap, apply encryption at rest and in transit.
• 11:53
  Host: Do we need FIPS-validated cryptography for CUI?
• 11:57
  Analyst: Yes, per NIST 800-171, FIPS validation is required for all controls.
• 12:03
  Host: What is the cost trade-off for private deployment?
• 12:06
  Analyst: At 500,000 tokens per day, private deployment breaks even within 6 to 12 months.
• 12:12
  Host: What if our usage is below that threshold?
• 12:15
  Analyst: Run a cost analysis before purchasing hardware; API may be cheaper.
• 12:19
  Host: How does the hardware sizing work for a 31B model?
• 12:23
  Analyst: It may need multiple GPUs; reference clusters use 128 GB unified memory per node.
• 12:30
  Host: Do we need to sign any new DFARS clauses?
• 12:33
  Analyst: No new clauses exist yet; consult your legal team to confirm flow-down.
• 12:38
  Host: What about the 60-day guidance deadline?
• 12:40
  Analyst: The Secretary must issue guidance within 60 days; monitor for that release.
• 12:45
  Host: How do we prepare for a potential 2027 bill?
• 12:49
  Analyst: Review the pending Section 1651; it could add new names to the list.
• 12:54
  Host: What if a model is not named but originates from a covered nation?
• 12:57
  Analyst: The statute does not generalize by nation; only named developers are covered.
• 13:02
  Host: What if we inadvertently use a covered model in a non-DoD contract?
• 13:07
  Analyst: It remains outside the statutory scope, but contractual flow-down may still apply.
• 13:12
  Host: Should we still audit that usage?
• 13:14
  Analyst: Yes, a proactive audit reduces risk across all operations.
• 13:18
  Host: What are the first concrete steps a contractor should take?
• 13:21
  Analyst: Start with an inventory of all AI models and their lineage.
• 13:25
  Host: Then map each to the 110 NIST controls?
• 13:29
  Analyst: Exactly; document how each model meets or fails the controls.
• 13:33
  Host: What about the system security plan?
• 13:35
  Analyst: Update it with the new AI environment, including hardware specs.
• 13:39
  Host: Do we need to document risk mitigation for any remaining covered models?
• 13:43
  Analyst: If a waiver is requested, capture all mitigation steps and risk assessments.
• 13:48
  Host: How do we avoid common documentation mistakes?
• 13:51
  Analyst: Ensure every entry references the exact model version and vendor.
• 13:55
  Host: What if the vendor changes the model after we deploy it?
• 13:59
  Analyst: Treat any new release as a new version; re-audit for coverage.
• 14:02
  Host: What questions do clients often ask about this law?
• 14:06
  Analyst: Do I need to remove the model from all systems, or just DoD workflows?
• 14:10
  Host: Do I need to get a waiver if I only use it for training?
• 14:13
  Analyst: Do I have to monitor for updates to the 1260H list?
• 14:17
  Host: Do I need to change my cloud provider to meet the FedRAMP baseline?
• 14:20
  Analyst: Do I need to re-validate encryption after a hardware upgrade?
• 14:24
  Host: What if we have a hybrid cloud environment?
• 14:26
  Analyst: Separate the DoD data path; keep the model in a compliant enclave.
• 14:31
  Host: How do we stay ahead of the pending Section 1651?
• 14:34
  Analyst: Track the legislative docket and engage with counsel for early guidance.
• 14:38
  Host: What is the impact on our procurement process?
• 14:41
  Analyst: Add a clause that requires the contractor to audit AI lineage.
• 14:45
  Host: Should we involve our compliance team early?
• 14:48
  Analyst: Yes, they can align the system security plan with the new requirements.
• 14:52
  Host: What is the role of the C3PAO in this context?
• 14:55
  Analyst: They assess the 110 controls but do not evaluate AI lineage.
• 15:00
  Host: Will the C3PAO need to re-audit after we replace a model?
• 15:04
  Analyst: They will verify the new model meets the controls and documentation.
• 15:08
  Host: What if we use an open-weight model that is not named?
• 15:11
  Analyst: It remains compliant with Section 1532, but you must still map it to NIST controls.
• 15:17
  Host: Are there any cost-saving tips for private deployment?
• 15:20
  Analyst: Use quantized models; they run on consumer GPUs and lower memory needs.
• 15:25
  Host: What if we need to process 5 million tokens daily?
• 15:29
  Analyst: Private deployment will be 60 to 80 percent cheaper than API spend.
• 15:34
  Host: What about the 773 to 1,391 Level 2 certifications trend?
• 15:39
  Analyst: It shows growing awareness; staying compliant positions you for more contracts.
• 15:44
  Host: Do we need to update our POA&M for AI changes?
• 15:47
  Analyst: Yes, add new action items and status updates for AI compliance.
• 15:52
  Host: What if we accidentally re-introduce a covered model?
• 15:54
  Analyst: Implement strict access controls and audit logs to detect such events.
• 15:59
  Host: How do we handle external consultants using our models?
• 16:03
  Analyst: Require them to sign attestations and confirm model compliance.
• 16:07
  Host: What about the 24/7 hybrid approach?
• 16:10
  Analyst: It keeps AI in a controlled environment but still requires model compliance.
• 16:15
  Host: Is there a benefit to using a model like Gemma 4?
• 16:17
  Analyst: It is open-weight, not covered by the statute, and under an Apache 2.0 license.
• 16:22
  Host: Do we need to purchase a license for Gemma 4?
• 16:25
  Analyst: The Apache 2.0 license is permissive; no purchase needed beyond hardware.
• 16:30
  Host: What if we need to fine-tune Gemma 4?
• 16:32
  Analyst: Fine-tuning is allowed; just document the process and data used.
• 16:36
  Host: Do we need to report fine-tuning to the Department?
• 16:39
  Analyst: No, unless the fine-tuning is part of DoD contract performance.
• 16:43
  Host: What if our model is from a 1260H-designated company?
• 16:47
  Analyst: Consult counsel before using it; it may be added to the statutory list.