Skip to content
Sign in
SWC 2026 FinalistGrand Final: Main Stage

Hardshell

Ex-DoD team building a data-centric security layer that detects poisoning and leakage in AI training datasets before models ship.

Charlottesville, Virginia, United States, North AmericaCybersecurity, AI training data security
Stage
Pre-seedAI integrated
Product
PrototypeNone known
Funding raised
$1.3Mdisclosed
Founded
20251 years ago
Founders
2Andrew, Hunter
Global score
4.3Grade DRank 76 of 151

Overview

Hardshell protects sensitive data used to train AI/ML systems against poisoning and leaks. Founded by Hunter Moore and Andrew Schoka, former DoD cyber/AI experts.

Traction

None known

Platform presented with a demo on request, with no public customer or contract; only prizes (about $170K).

Market

AI security (generative AI / training data)

As organisations train and fine-tune models on sensitive data, training sets become an attack surface (poisoning) and a leakage risk. AI regulation and defence procurement are pushing for assurance of training data. Buyers are CISOs, AI platform leads and defence programme offices, often via pilots. Large cybersecurity vendors move quickly into the space.

Addressable market

2025
$8.7B
2030
$28B
+26.5% a year

AI security (generative AI / training data): $8.65bn in 2025, +26.5%/year, $28.02bn in 2030 (MarketsandMarkets).

Customers

Defence and government AI programmes, enterprises training models on sensitive data, data marketplaces

Who signs the cheque

CISOs, heads of ML platforms, defence programme offices

Geographies

  • United States

Drivers

  • Growth of model training on proprietary data
  • AI regulation and governance requirements
  • Defence demand for trusted AI

Headwinds

  • Fast entry of large cyber vendors
  • No customers yet
  • Long government sales cycles

The generative AI cybersecurity market grows from $8.65bn in 2025 to $35.5bn in 2031 (+26.5%/year) according to MarketsandMarkets.

Scalability 8/10. AI security software deployable worldwide, with demand driven by AI regulation; constraint: no customer yet.

Market source: contentenginellc.com

Technology

Category

Cybersecurity – AI/ML data security

AI integratedMachine learning / predictive

Maturity

  1. R&D
  2. Prototype
  3. Pilot
  4. On the market
  5. Scaling

prototype

How it works

A data-centric platform scans training datasets to detect vulnerabilities, poisoning and leakage. It is offered with demos on request.

Moat

Team expertise from DoD AI security; product moat unproven.

Differentiation

Focus on training data rather than model or prompt security.

Integration points

  • ML pipelines and data lakes
  • Data marketplaces
  • AI reliability/uncertainty tools
  • Pentest and CTEM platforms

Depends on

  • Access to customer data environments
  • Government certifications
  • ML framework integration

Use of AI

Platform that detects poisoning, vulnerabilities and leaks in AI/ML model training data.

Product in detail

"Data-centric" AI security platform that detects vulnerabilities, poisoning and leaks in training datasets.

Business

Model
Enterprise/government software licences (not disclosed).
Revenue logic
Per-dataset or per-platform subscription (assumption).
Go-to-market
Defence and veteran ecosystem, prizes and demos.

Next milestones

  1. First paid pilot
  2. Government contract (SBIR)
  3. Seed round

Key risks

  • No customers
  • Competition from incumbents
  • Narrow window

Opportunities

4 found in the network

Concrete moves for Hardshell inside the network: who to sell to, sell through, build with or raise from, each with the reasoning, the evidence and the first step.

  • Customers 2
  • Distribution 1
  • Co-development 1

Customers2

Who could buy, or sell through

Certify Cromodata's clinical datasets against leakage and poisoning

Cromodata sells de-identified clinical data for AI training; buyers will ask for assurance against re-identification and contamination. Hardshell's scans give Cromodata a trust label and Hardshell a commercial, non-defence reference.

40%confidence

Evidence from the records

  • “Argentine healthtech that runs a marketplace for de-identified clinical data for research and AI training”
  • “"Data-centric" AI security platform that detects vulnerabilities, poisoning and leaks in training datasets.”
If it works
A paid dataset-audit pilot and a 'verified dataset' feature.
First step
Offer Cromodata a free scan of one imaging dataset.

Secure training corpora for Deepentix's regulated-industry SLMs

Deepentix trains vertical small language models on internal documents in pharma and banking, where leakage is critical. Hardshell can scan those corpora; Deepentix gains a security proof point for European buyers.

25%confidence

Evidence from the records

  • “"CLAIM Graph" and vertical small language models (SLMs) providing a deterministic chain of evidence traceable back to the sources, at 15–45x lower cost.”
  • “Platform that detects poisoning, vulnerabilities and leaks in AI/ML model training data.”
If it works
An integration and a first European pilot.
First step
Share a technical brief with Deepentix on integration with its training pipeline.

Distribution1

A channel or customer base to reach together

Resell AI-data audits through Pwn&Patch's security clients

Pwn&Patch claims 300+ customers and sells via AWS Marketplace; adding AI-data audits extends its offer. Hardshell gets a channel in North Africa and Europe.

20%confidence

Evidence from the records

  • “Oktoboot platform available on AWS Marketplace; the website claims more than 300 customers and 40+ consultants (self-reported figures).”
  • “Platform presented with a demo on request, with no public customer or contract; only prizes (about $170K).”
If it works
A reseller agreement and first international customers.
First step
Propose a reseller pilot to Pwn&Patch.

Co-development1

An integration or joint product

Bundle data security with model reliability for defence AI assurance

Neurath measures model uncertainty and blind spots with defence-grade compliance, and Hardshell secures the training data. Together they cover data-to-model assurance for defence AI buyers.

30%confidence

Evidence from the records

  • “Uncertainty quantification: Trireme SDK and Gauloi Kubernetes platform deployed on-premises/VPC (SOC 2, ISO 27001, NIST 800-171, CMMC compliance).”
  • “Protecting AI training data against poisoning: a new approach to an emerging risk.”
If it works
A joint proposal to a defence AI programme.
First step
Arrange a call to define a joint assurance workflow.

Analyst reading of 21 Nov 2026, from the public record of the network. Method: analyst-ai.

Investors that fit

Pre-seed fund writing first cheques into software companies, judging business model, unit economics and founder hustle.

US pre-seed AI training-data security — software, US, first-cheque stage; strong syndicate already formed.

60%confidence

Evidence from the records

  • “"Pre-seed"”
  • “AI training data security”
  • “Platform presented with a demo on request, with no public customer or contract”
Intro path
Founder-to-founder intro via Baniql (network portfolio company of Hustle Fund).
First step
Ask Hardshell's investors (Cleo Capital, Front Porch) to forward.

US federal department funding dual-use startups through ESTCP and xTechSBIR, as received by waste-to-energy company Enexor.

AI training-data security founded by former DoD cyber/AI experts: a direct defence need (data poisoning), US-based, pre-seed.

55%confidence

Evidence from the records

  • “AI training data security”
  • “Founded by Hunter Moore and Andrew Schoka, former DoD cyber/AI experts”
Intro path
VTC Ventures (Hardshell investor).
First step
Respond to the next DoD SBIR topic on AI assurance.

Zero-equity nonprofit accelerator (since 2009) running corporate-funded programmes in Boston, Dallas, Jerusalem, Lausanne and London.

US AI-training-data security at pre-seed — Cybersecurity track.

45%confidence

Evidence from the records

  • “"Cybersecurity"”
  • “AI training data security”
  • “Platform presented with a demo on request, with no public customer or contract”
Intro path
Founder-to-founder intro via Monte Caldera Technologies Inc. (network portfolio company of MassChallenge).
First step
Apply.

NATO's Defence Innovation Accelerator for the North Atlantic, supporting pre-seed and seed dual-use deep tech in Europe and North America.

US AI training-data security — AI assurance.

40%confidence

Evidence from the records

  • “North America”
  • “AI training data security”
  • “Platform presented with a demo on request, with no public customer or contract”
Intro path
Founder-to-founder intro via Astrolight (network portfolio company of NATO DIANA).
First step
Apply.

Israeli government agency funding innovation; its Tnufa ideation programme gives up to NIS 200,000 (85% of budget), repaid via royalties.

US AI-security — cybersecurity; needs Israeli R&D entity.

10%confidence

Evidence from the records

  • “Israel”
  • “AI training data security”
  • “Platform presented with a demo on request, with no public customer or contract”
Intro path
Founder-to-founder intro via EVR Motors Ltd. (network portfolio company of Israel Innovation Authority).
First step
Low priority.

Investment analysis

4.3D

Grade D

Rank 76 of 151

Disruption6.0/10
Business model2.0/10
Market6.0/10
Risk(lower is better)8.0/10

Global score = mean of disruption, model, market and (11 − risk). Last updated 21 Nov 2026.

Promising thesis but pre-commercial: too early.

Disruption

6/10

Protecting AI training data against poisoning: a new approach to an emerging risk.

Business model

2/10

Demo on request, no public customer or contract.

Market

6/10

TAM $8.65bn → base 5; scalability 8/10 (+1); no international presence: no public customer → 6/10.

Risk

8/10

Lower is better

High risk: prototype, fast competition from large cyber players, US$1.3M raised.

Strengths

  • Team from the DoD
  • Promising AI security theme

Risks

  • No customers
  • Narrow competitive window

Funding

$1.3M, disclosed
  1. 2026-02-13

    pre-seed$1.1M

    VTC Ventures, Front Porch Venture Partners, Blu Ventures, MoJo Ventures, Not Yet Ventures, Black Prism Capital Partners, CAV Angels, angels (Cleo Capital also listed on the website)

  2. 2026-04-23

    prize/grant$100K

    Veteran Startup Showcase (Service First Foundation)

  3. 2026-06-20

    prize$25K

    Next Level: Growth Exchange (Greater Washington Partnership)

  4. 2026-08-24

    prize$25K

    VetsinTech Pitch Contest (J.P. Morgan)

  5. 2026-09-17

    prize$20K

    Virginia Cup (Startup World Cup qualifier)

Amount used in the study. $1.1M pre-seed + four cash prizes totaling $170K.

Note. ~$1.1M (pre-seed) + ~$170K in prizes

verified: $1.1M pre-seed (Feb. 2026), Veteran Startup Showcase ($100K) and VetsinTech ($25K) prizes confirmed; Virginia Cup win (SWC qualifier, 17 Sept. 2026) not found in an independent source, regional field left empty.

Team

  • Andrew Schoka

    Co-founder & CEO

    Former US Army cyber warfare officer (Cyber Command, USSOCOM); MBA UVA Darden, MS in cybersecurity Georgia Tech.

  • Hunter Moore

    Founder & CTO

    PhD in systems engineering (UVA); 5+ years of AI security consulting for the DoD (JAIC/CDAO).

Sources

Identification confidence
medium
Founding year: 2025 (medium confidence)
Founded in 2025 (Charlottesville, VA) according to PitchBook and press releases. pitchbook.com
  1. 1cioinfluence.com/security/hardshell-raises-1-1m-to-deliver-the-data-centric-
  2. 2news.darden.virginia.edu/2025/07/17/qa-protecting-ai-from-the-inside-out-with-startu
  3. 3caplight.com/company/hardshell
  4. 4norfolkdailynews.com/online_features/press_releases/enterprise-data-layer-assura
  5. 5vabio.org/?p=7491

From the open study Startup World Cup 2026 by TIBOK, based on public sources, verified in November 2026. Scores are an analyst's judgment on an explicit grid, not the jury's.

Wall

Wall

No posts yet