Research policy · v1.0 · effective 31 May 2026

What we research — and what we explicitly don't.

DCS Labs is a research organisation. The 72-feature agenda spans government, defence, medical, space, financial, climate, disaster, education, and agriculture. Most are sensitive. This page documents — category by category — what we research, the legal frameworks we operate under, and the things we will not do without proper partnerships, certifications, or authorisations.

Document: DCS-LABS-POLICY-1.0 Effective: 31 May 2026 Review cadence: quarterly Owner: DCS AI Technologies L.L.C License: 1624450
Mission

Why DCS Labs does this research.

A handful of US labs publish all the foundational AI work. The rest of the world adopts it years later, often without the sovereignty, audit, or domain-fit that local institutions need. DCS Labs exists to close that gap. The 72-feature agenda is the set of capabilities we believe sovereign AI infrastructure should have — built openly, with proofs, in public.

01

Sovereignty by default

If a customer's data leaves their jurisdiction, it loses legal protection. Most AI infrastructure assumes US-controlled cloud. We research the opposite: cryptographically-verifiable sovereignty as a built-in property, not an add-on.

02

Auditability over hype

Every action — every receipt, model decision, agent move — should be independently verifiable years later. The R-Series cryptography (R+1/+2/+3/+4) is research toward making that practical at production scale.

03

Open standards · closed weapons

Cryptography, receipts, agent protocols → published, MIT-licensed, peer-reviewable. Weaponizable applications → not built without sovereign partnerships, IRB-style review, and explicit governmental authorisation.

04

Research as a public good

Every paper, prototype, dataset, and benchmark we produce is published under permissive licenses (MIT / CC-BY-4.0). The 9-piece moat is in code; the standards are in specs; the agenda is on this page. Build with us, fork us, audit us.

Per-category legal framing

What we can — and cannot — research, by domain.

Each of the 9 sensitive categories below has its own legal frame. Research mode = whitepapers, prototypes, public demos, datasets, benchmarks. Operational mode = deployment into the regulated environment, requiring partnership, certification, and authorisation we do not currently hold.

Domain 01

Government

Research mode
Sample features:
#39 · #47 · #67

What we research

Sovereign-AI reference architectures for government use — air-gapped deployments, full audit chains, multi-ministry coordination patterns, citizen-data protection by cryptographic erasure. Published as whitepapers + reference code.

What we do not do

We do not operate government systems. We do not have classified-information clearances. We do not bid on tenders for systems handling national security data without a partnership-of-record and a defence-ministry-issued authorisation.

Legal frame

UAE Federal Decree-Law 45/2021 (PDPL) India DPDP Act 2023 GDPR (EU residents) Local government procurement law (where engaged)

Domain 02

Defence

Research mode · partnership-only
Sample features:
#8 · #16 · #20 · #21 · #22 · #41 · #46 · #50

What we research

Dual-use AI primitives that have a publishable, peer-reviewable, non-weaponized research surface — secure communication protocols, multi-agent coordination patterns, supply-chain optimisation models, threat-detection benchmarks. All published openly.

What we do not do

  • No autonomous weapon systems. Ever.
  • No targeting, kill-chain, or strike-decision components.
  • No deployment into operational defence networks without a sovereign defence partnership-of-record (DRDO / equivalent) and an explicit authorisation from the relevant defence ministry.
  • No export of restricted-list technologies outside ITAR / EAR / Wassenaar permissions.

Legal frame

US ITAR 22 CFR §120-130 US EAR 15 CFR §730-774 Wassenaar Arrangement (dual-use list) India SCOMET list (DGFT) UAE export-control law (Federal Decree-Law 13/2007)

Domain 03

Medical / Healthcare

Research mode · not diagnostic
Sample features:
#25 · #26 · #27 · #52 · #53 · #54 · #55 · #72

What we research

Privacy-preserving healthcare AI — disease-prediction models on anonymised data, medical-image analysis benchmarks, mental-health support agent architectures with sovereign data guarantees, drug-discovery acceleration prototypes. Output: papers, datasets, reference code.

What we do not do

  • No diagnostic output for clinical use. No "this image shows cancer" classifications presented to a patient or clinician as actionable medical advice.
  • No FDA / CE / CDSCO-cleared product claims until we have the relevant clearance for the relevant product in the relevant jurisdiction.
  • No mental-health crisis intervention as a primary care channel. Our research includes detection prototypes; any deployment is referral-only, with a licensed-professional human-in-the-loop, never autonomous response to a crisis.
  • No PHI processing in any form without a signed BAA (US) / DPA (EU) / equivalent.

Legal frame

HIPAA (US, with BAA) FDA 21 CFR §820 / SaMD guidance EU MDR 2017/745 India CDSCO Medical Devices Rules 2017 GDPR Art 9 (sensitive data) UAE Federal Law 2/2019 (health data)

Domain 04

Space

Research mode · partnership-only
Sample features:
#7 · #9 · #14 · #15 · #17–19 · #23 · #24 · #40 · #42 · #44 · #45 · #48 · #49 · #51 · #68

What we research

Edge-AI architectures for space (low-power agents, intermittent connectivity), on-chain provenance for satellite imagery, mission-control assistant prototypes, multi-planet agent coordination protocols, lunar-resource-mapping AI, space-weather prediction networks. Open papers + reference implementations.

What we do not do

  • We do not operate space hardware or fly anything. Our research is the software + protocols that space agencies (ISRO, NASA, ESA, partners) may choose to evaluate.
  • No claims of ISRO / NASA / ESA partnership are made unless a formal MoU is signed and publishable.
  • No defense-adjacent satellite imagery analysis sold to private buyers (see ITAR / EAR above).

Legal frame

UN Outer Space Treaty 1967 US Remote Sensing Act / NOAA licensing India Remote Sensing Data Policy 2011 ITAR USML Cat XV (spacecraft + satellites) UAE Space Activities Law (Federal Law 12/2019)

Domain 05

Financial

Research mode · not advice
Sample features:
#34 · #35 · #63 · #64

What we research

Fraud-detection patterns for banking + UPI, alternative credit-scoring on non-traditional data (with consent), insurance-fraud detection benchmarks. Models published as research; not personalised financial advice.

What we do not do

  • We are not a SEBI / RBI / FCA / SEC / DFSA registered financial-services advisor. Nothing we publish is investment, lending, or insurance advice.
  • We do not process payments as a payment institution. (Our products use Stripe / SEPA / USDC; we are a SaaS, not a money-services business.)
  • No credit scoring as a service sold to lenders until we have the relevant credit-reporting authorisations (RBI / equivalent) in the relevant jurisdiction.
  • No use of synthetic identity for fraud-detection bypass research without explicit consortium-of-banks consent.

Legal frame

RBI Master Directions (India) UPI Procedural Guidelines (NPCI) PCI-DSS EU PSD2 UAE CBUAE regulations SEC / SEBI / FCA depending on customer jurisdiction

Domain 06

Climate

Research mode
Sample features:
#30 · #58 · #59 · #60

What we research

Carbon-credit verification with satellite + on-chain proofs, air-quality prediction + health alerts, glacier + sea-level monitoring, climate-risk prediction for farmers. Open-source benchmarks against public datasets (NASA EOSDIS, ESA Copernicus, India ISRO Bhuvan).

What we do not do

  • We do not certify carbon credits ourselves. Our research is verification infrastructure that recognised registries (Verra / Gold Standard) may adopt.
  • No catastrophe-bond pricing or insurance products without proper regulator engagement.

Legal frame

UN Paris Agreement Art 6 Verra VCS / Gold Standard methodologies (as a reference, not certifying body) India BEE / MoEFCC notifications EU EU ETS

Domain 07

Disaster management

Research mode · ndma-aligned
Sample features:
#31 · #32 · #33 · #61 · #62

What we research

Disaster early-warning models, flood + cyclone prediction, wildfire detection from satellite/drone imagery, earthquake-pattern research, emergency-response coordination protocols. Aligned with India NDMA + UN Sendai Framework references.

What we do not do

  • We do not issue disaster warnings directly to populations. Warnings are issued by NDMA / IMD / national agencies; our research feeds them, not replaces them.
  • No solo-actor emergency-response automation — humans-in-loop and licensed authorities always remain in the loop.

Legal frame

India Disaster Management Act 2005 UN Sendai Framework 2015-2030 WMO Common Alerting Protocol (CAP) FEMA IPAWS (US, where applicable)

Domain 08

Agriculture

Research mode
Sample features:
#28 · #29 · #56 · #57

What we research

Crop-health monitoring from satellite/drone imagery, precision-farming agent recommendations, soil-health + fertiliser optimisation, supply-chain optimisation models for FCI-scale procurement. Open benchmarks; pilots with farmer-cooperatives where invited.

What we do not do

  • No agronomic prescriptions sold without local agriculture-extension partnership (KVK / equivalent).
  • No GMO / seed-engineering research.
  • No data extraction from farmers without explicit, plain-language consent + revenue share.

Legal frame

India Plant Quarantine Order 2003 PPV&FR Act 2001 (farmer rights) EU CAP regulations FAO Voluntary Guidelines on responsible tenure

Domain 09

Education

Research mode · age-appropriate
Sample features:
#36 · #37 · #65 · #66

What we research

Personalised-learning agent architectures, skill-gap analysis models, NEET/JEE coaching agent prototypes, vernacular-language learning agents. All COPPA-aware, parent-consent first for minors.

What we do not do

  • No data collection on minors without parental consent (COPPA-style framework even outside US jurisdictions).
  • No autonomous grading / placement decisions presented as the school's official decision — always teacher-in-loop.
  • No emotional-analysis-of-students sold to administrators without published research showing harm-free use.

Legal frame

COPPA (US, under-13) India DPDP 2023 (children's data) GDPR Art 8 (consent age) UNESCO AI in Education recommendations

Hard limits · what we will never build

Things outside our research scope — permanently.

Some research areas are technically possible but ethically and legally off-limits. These will never enter our roadmap. If you ask for them, the answer is no.

Forbidden 01

Autonomous weapon systems

Targeting, kill-chain, autonomous strike decisions, lethal-force-without-human-in-loop. Not researched, not prototyped, not consulted on.

Forbidden 02

Mass surveillance toolchains

Face-recognition + cross-referencing-without-warrant pipelines. Population-scale tracking. Social-credit scoring.

Forbidden 03

Biometric inference at scale

Gender / sexuality / political-affiliation / health-status inference from face or voice. No exceptions.

Forbidden 04

Undisclosed AI in deceptive contexts

Generative voice/text representing a real person without consent. Synthetic media in elections. Romance-scam AI.

Forbidden 05

Bio-weapon / chemical-weapon precursor research

Any AI assistance to bio / chem / nuclear weapons design. Includes "academic" curiosity prompts on this topic.

Forbidden 06

Child-safety violations

CSAM generation, detection-bypass research, age-verification-bypass research. Reported to authorities if asked.

Forbidden 07

Critical-infrastructure attack research

Offensive cyber-attack tooling for power-grid, water, hospital, transport, financial infrastructure. We may research defence; never offence.

Forbidden 08

Election-influence operations

Voter-targeting automation, disinformation generation, deep-fake political ads. Includes paid consultancy on this topic.

Internal review process

How research items pass our internal review.

Every feature on the 72-item roadmap that touches one of the 9 sensitive categories above goes through this 5-step review before any code is written. Items in the "hard limits" list above are auto-rejected at step 1.

Step 1 · Category classification

Each proposed feature is classified by domain (Government / Defence / Medical / Space / Financial / Climate / Disaster / Agriculture / Education / Core). The forbidden list is checked first — any match auto-rejects.

Step 2 · Legal-framework mapping

The relevant statutes are listed (e.g. HIPAA + FDA SaMD for medical, ITAR + EAR for defence). If we lack the relevant authorisation, the feature is downgraded to "research-mode only" — no operational deployment.

Step 3 · Harm model

What's the worst-case misuse? Who's the affected population? What harm-mitigation primitives must be in the design (consent, opt-out, audit, kill-switch)?

Step 4 · Disclosure plan

How will we be honest about what's built vs. theoretical? What status pill (Shipped / Partial / Roadmap) is accurate? What flags ship default-OFF?

Step 5 · Reversibility

If after launch we discover unintended harm, can we roll back? Cryptographic erasure, kill-switches, model unloads, and consent revocation must work — proven before launch, not after.

Cross-cutting frameworks we comply with

The umbrella legal stack.

Beyond category-specific law, DCS Labs operates under these umbrella frameworks for every product and every customer.

DP

Data protection

GDPR (EU + UK), India DPDP 2023, UAE Federal Decree-Law 45/2021, US state laws (CCPA / CPRA / VCDPA), Brazil LGPD. DPA template at dcsai.ai/dpa.

XC

Export control

US ITAR + EAR, Wassenaar Arrangement, India SCOMET (DGFT), UAE export-control law. No dual-use technology export without licence verification.

IP

Intellectual property

Code under MIT/Apache-2. Standards under CC-BY-4.0. We respect third-party IP; takedown process at [email protected].

AI

AI-specific regulation

EU AI Act risk-tier compliance (high-risk = research-mode only). NIST AI RMF reference. India NITI Aayog principles. UAE AI Charter. Voluntary G7 Hiroshima principles.

SE

Security

Responsible-disclosure: /security. Bug-bounty for receipt-chain + R-Series vulnerabilities. PGP key at /security/pgp.txt.

AC

Anti-corruption

US FCPA, UK Bribery Act 2010, India Prevention of Corruption Act, UAE Federal Decree-Law 31/2021. Zero tolerance, every customer.

Have a question? Want to push back?

If you think a feature on our roadmap should not exist, tell us. If you think we're being too cautious in a category that matters to you, also tell us. We update this policy quarterly + every time a partner asks a question we hadn't answered.