Jeremy Samuelson at Integrated Quantum Technologies shares his insights in this Perspectives article for AZoQuantum, arguing that “post-quantum AI infrastructure” is not just industry shorthand but a practical business concern. As quantum computing moves closer to real-world impact, Samuelson makes the case that enterprises should focus less on responding to a far-off risk and more on rethinking how sensitive data moves through AI systems right now, before long-held assumptions are put to the test.
Why “Post-Quantum AI Infrastructure” Is More Than a Buzzword
In enterprise AI, “post-quantum” should not be treated as a slogan. It is really about how long your security assumptions remain valid. AI systems concentrate large volumes of sensitive data, operational workflows, and model IP. So the question is not whether quantum computing suddenly changes everything overnight. The question is whether the way you handle raw data today will still look defensible over the lifetime of the models and data assets you are creating now.
For that reason, post-quantum readiness in AI is less about making louder claims about future threats and more about reducing structural exposure in the present. If an AI pipeline depends on raw sensitive data being copied, moved, or broadly exposed across downstream systems, then the organization is carrying a larger long-term risk surface than it needs to.
That is where VEIL fits. VEIL is designed to apply the encoder inside a trusted source environment and allow downstream machine learning workflows to operate on latent representations rather than raw sensitive inputs. The point is not to dramatize quantum risk. The point is to build AI systems whose privacy model is less dependent on the future strength of perimeter defenses alone, and more rooted in architectural data minimization from the start.
Image Credit: Phonlamai Photo/Shutterstock.com
Rethinking Data Security Inside the AI Pipeline
For years, enterprise AI security has focused on the front end of the pipeline: collection, storage, access control, encryption, and governance. Those controls still matter. But AI changes the scale and persistence of the problem. Once raw data are widely available across training, fine-tuning, scoring, and downstream integrations, the pipeline itself becomes part of the attack surface.
That is why I think the next phase of AI security has to be architectural. The goal should not be only to protect sensitive data wherever they travel. It should also be to ask whether they need to travel in raw form at all. In other words, the deeper question is how much sensitive information downstream systems actually need to see.
VEIL is built around that principle. It is a privacy-preserving ML infrastructure layer that transforms raw inputs into deterministic latent encodings inside the customer’s own environment, so that downstream training and inference can operate on reduced-risk representations instead of the original sensitive records. In the Snowflake implementation, for example, processing stays inside the consumer account, with no external network calls and no provider-side runtime access to customer data. That is a materially different posture from treating privacy as a bolt-on after the pipeline is already exposed.
I would also frame the comparison carefully. Homomorphic encryption and differential privacy are important approaches, and in the right settings they can be valuable. But many enterprise teams find that those methods can introduce real trade-offs in complexity, latency, or utility depending on the workload. VEIL is aimed at a different design point: enforcing privacy through where computation happens and what leaves the trusted environment, rather than assuming that every downstream system must handle raw sensitive data directly.
What Enterprise Leaders Should Be Doing Now
Enterprise leaders do not need to make speculative bets on quantum timelines, but they do need to review the architecture of their AI data path with much more discipline. The practical questions are straightforward: where do raw sensitive inputs move, which systems actually need them in original form, what crosses the trust boundary, and what would still be exposed if a downstream training or inference environment were compromised?
That changes the vendor conversation. CISOs, CIOs, and AI leaders should ask not only whether a solution is encrypted and compliant, but whether it minimizes raw-data exposure by design. They should ask where encoding happens, whether raw inputs or gradients ever leave the trusted environment, whether a decoder exists anywhere in the deployed stack, and what the operational trade-offs are.
For organizations handling highly sensitive data, VEIL is meant to answer those questions at the infrastructure layer. It keeps raw data and the encoder inside the trusted source environment, exports only latent representations for downstream ML operations, and is designed so those representations are structurally non-invertible rather than merely hidden behind policy or perimeter controls. The broader lesson is that enterprise AI strategy now has to include architecture strategy. The organizations that get this right will be the ones that can scale AI across jurisdictions and high-sensitivity use cases without treating privacy, performance, and innovation as mutually exclusive.
Download the PDF of this article here
About the Speaker
Samuelson is a data scientist and mathematician by trade with nearly 2 decades of experience.
Prior to Integrated Quantum Technologies, Samuelson served as the Principal Data and AI Scientist for Digital Identity Engineering at Equifax. Prior to Equifax, he held senior AI leadership roles at Mastercard and VICI Capital Partners. Additionally, he teaches graduate-level executive programs in AI and management at leading institutions, including John Hopkins and the University of Austin.
About Integrated Quantum Technologies
Integrated Quantum Technologies is a developer of quantum-ready AI infrastructure for global organizations that handle highly sensitive data. The Company’s primary goals are to address security risks, including post-quantum threats, growing compute demands, and the increasing complexity of deploying AI at scale and across jurisdictions.
Integrated Quantum Technologies has developed AIQu VEIL™ (Vector-Encoded Information Layer) which uses Information Compressive Anonymization (ICA) to convert structured data into compact, non-invertible vector representations. In other words, it removes sensitive information from data going into AI model pipelines while preserving the utility.
VEIL™ Provides
- End-to-end PHI protection across the ML lifecycle
- Persistent privacy from ingestion through inference
- High-performance vector representations for AI
- Built-in alignment with HIPAA and global data regulations
- Minimal workflow disruption for internal teams
- Future-proof protection for emerging data types