Issued on behalf of Integrated Cyber Solutions Inc.
As Chinese state-sponsored actors weaponize frontier AI against U.S. enterprises and Washington reframes data exposure as a national security problem, Integrated Cyber Solutions Inc. (dba Integrated Quantum Technologies) has published an updated white paper reporting 95%+ compression of sensitive data — removing it from the AI attack surface entirely while maintaining model performance across healthcare, financial services and enterprise-scale environments.
Equity Insider — In November 2025, Anthropic disclosed that Chinese state-sponsored actors had used its Claude model to run a largely automated cyberespionage campaign across roughly thirty targets, with the AI performing 80 to 90 percent of the operational work. [9] Five months later, in April 2026, the White House Office of Science and Technology Policy issued a memo warning that foreign entities, primarily based in China, are conducting industrial-scale campaigns against U.S. frontier AI systems. [10] On May 18, 2026, the Council on Foreign Relations published an assessment titled “The Security Foundations Beneath America’s AI Ambitions Are Cracking.” [11] In the span of six months, enterprise data exposure to AI systems has stopped being a corporate IT problem and started being a national security problem.
That reframing matters at the boardroom level, because every enterprise running a serious AI program eventually runs into the same wall. The data that would make their models genuinely useful — patient records, transaction history, claims data, internal financial filings, regulated images — is also the data their legal and security teams will not let near a model pipeline. So they ship synthetic substitutes, or they aggressively anonymize, or they encrypt and pay the latency cost, or they just narrow the project until the data risk goes away. Whichever path they pick, the model that ships at the end is a weaker version of what was actually possible. And in a threat environment where state-sponsored actors are now using AI itself to harvest that data, the cost of leaving it exposed has gone up sharply.
Also Read: CIO Influence Interview with Kyle Wickert, Field CTO at AlgoSec
Stay ahead of the AI security and post-quantum stories investors are watching. Sign up for the Equity Insider newsletter for ongoing coverage.
That is the bottleneck. And it is the bottleneck that a Canadian-listed company called Integrated Cyber Solutions Inc. (CSE: ICS| OTCQB: IGCRF | FSE: Y4G), which now operates publicly as Integrated Quantum Technologies (“Integrated Quantum,” “IQT,” or the “Company”), has been quietly working to dissolve.
On May 26, 2026, the Company released an expanded version of its white paper on VEIL™, its commercial product for privacy-preserving machine learning. [1] The paper, authored by Jeremy J. Samuelson, EVP, Artificial Intelligence & Innovation, is titled “Informationally Compressive Anonymization: Non-Degrading Sensitive Input Protection for Privacy-Preserving Supervised Machine Learning,” and is available here. The title alone is the thesis: the central claim of the work is that an enterprise can compress sensitive inputs by between approximately 95% and 99.96%, demonstrate resilience against reconstruction and attribute inference attacks under the testing conditions described, and at the same time match — or in some cases beat — the predictive performance of a model trained on the raw data. [1]
If that holds up under real-world deployment, it is a meaningful claim. Privacy-preserving ML has historically been a graveyard of “almost” solutions. Differential privacy degrades accuracy by injecting noise. Homomorphic encryption multiplies computational cost. Federated learning still leaks gradients under the right attack. In each case, the engineer running the project has had to decide which tax to pay: the accuracy tax, the compute tax, or the security tax. The pitch in the Samuelson paper is that VEIL™ materially narrows that trilemma by removing sensitive information before it ever enters the ML pipeline, rather than trying to protect it once it gets there.
What the Paper Actually Shows
The updated paper is not a marketing one-pager. It evaluates VEIL™ across multiple supervised machine learning tasks and datasets, in image recognition, financial services, healthcare, regression modeling and large-scale enterprise data environments. The benchmark and enterprise datasets it tests include MNIST, Fashion-MNIST, Ames Housing, YearPredictionMSD, Home Credit Default Risk, Default of Credit Card Clients, CBIS-DDSM medical imaging data and the E2006 financial filings dataset. [1] That is a deliberately wide net — toy benchmarks alongside enterprise-grade data — because the company is making a generalizability argument, not a single-benchmark argument.
The two headline numbers are worth restating. Reported compression levels across the evaluated datasets and machine learning tasks ranged from approximately 95% to 99.96%, depending on the dataset, dimensionality and model architecture utilized. [1] And in each evaluation, VEIL™ maintained predictive utility comparable to and/or exceeding baseline raw-data model performance. [1] The combination matters. Either one in isolation would be unremarkable: compression without utility is just lossy data, and utility without compression is just regular ML on regular data. The claim is that you get both.
The paper also benchmarks VEIL™ directly against the two privacy-preserving approaches most often discussed in enterprise procurement conversations: Differential Privacy and Homomorphic Encryption. Both are associated with predictive performance trade-offs in addition to computational overhead, privacy-budget management requirements and ciphertext expansion characteristics under certain implementations and testing conditions. [1] Under the evaluated testing conditions described in the paper, VEIL™ outperformed Differential Privacy across the reported attack simulations — simulations that include reconstruction attacks and attribute inference analyses intended to assess resilience under various threat scenarios and attacker assumptions. [1]
The Company is careful, to its credit, about overclaiming. The paper notes that in certain enterprise deployment scenarios involving vulnerabilities elsewhere in a system environment — leaked sensitive indices, external data correlation — VEIL™ may still permit limited sensitive information leakage under specific adversarial conditions. [1] That is the honest version of the claim. The findings, performance observations and comparative analyses contained in the paper are based on internal research, simulations, validation studies, datasets, configurations and assumptions utilized by the Company and the paper’s author; results may not be indicative of performance in all commercial deployments. [1]
An external endorsement also accompanies the release. Dr. Mohammad Tayebi, Assistant Professor in the School of Computing Science at Simon Fraser University, who was referenced in the Company’s original white paper announcement, supports and endorses the updated paper. The Company has disclosed that Dr. Tayebi has no affiliation with Integrated Quantum and has received no compensation from the Company in connection with the endorsement, the white paper or the underlying research. [1]
Why the Compression Number Matters Beyond Privacy
There is a second story tucked inside the headline. The Company believes that the ability to materially reduce dataset size while preserving model utility may have broader implications for enterprise AI infrastructure efficiency, including potential reductions in storage, transfer and computational requirements associated with certain machine learning workflows. [1]
Put plainly: if an enterprise can shrink the information footprint of its sensitive training data by 95%-plus and still get the same model performance, the downstream implications for compute and storage envelopes may be material. The Company itself frames this as “potential reductions in storage, transfer and computational requirements associated with certain machine learning workflows.” [1] The privacy benefit is the on-ramp, but the infrastructure-cost benefit is what could keep VEIL™ on a finance team’s radar after the security team is already convinced. Enterprise AI has become a budget line item large enough that even modest reductions in compute and storage translate into meaningful savings.
Samuelson framed it this way: “Our research continues to support the view that informational compression and architectural isolation may provide a viable framework for privacy-preserving machine learning without requiring the substantial computational overhead commonly associated with certain existing approaches. We also believe the compression characteristics demonstrated in the paper could have meaningful implications for enterprise AI efficiency and infrastructure optimization in certain deployment scenarios.” [1]
The Public-Market Read-Across: A Sector Repricing in Real Time
The capital markets have not been subtle about what they think of companies positioned at the intersection of AI security and enterprise data protection. Four public names — each operating at a different layer of the same broad stack — give a sense of how investors are paying for this thesis right now.
Catch more CIO Insights: The CIO as a Value Creator: Moving Beyond Cost Centers to Revenue Drivers
[To share your insights with us, please write to psen@itechseries.com ]
