Skip to content
Trust & governance

Command stays sovereign.

VISHVAKSHA is a decision-support and intelligence-fusion platform. Consequential decisions remain under accountable human command, by architecture, not by policy alone.

What the platform must not do

Independently authorise lethal action

Directly release or control weapons

Bypass command approvals

Treat a language-model response as operational truth

Allow unrestricted autonomous agents to control military assets

Process classified information through unapproved external APIs

Hide uncertainty, provenance or contradictory evidence

AI risk tiers

Automation boundaries scale with consequence. Nothing above tier three moves without human review, and tier four is never delegated to general-purpose AI.

TierExampleAutomation boundary
0Search, translation, summarisationAutomatic, with full logging
1Maintenance recommendationHuman acknowledgement required
2Logistics prioritisationAuthorised approval required
3High-consequence operational recommendationMulti-level human review
4Lethal or irreversible actionNot delegated to general-purpose AI

Every AI output carries

Result and supporting evidence
Contradictory evidence, where it exists
Data timestamps and classification
Confidence and uncertainty
Model identity, version and limitations
Required reviewer and audit reference
Security architecture

Zero trust, at every boundary.

Access decisions combine identity, role, mission, classification, compartment, device trust and workflow state, evaluated fresh on every request, not cached indefinitely.

Mutual TLSHardware-backed identityShort-lived credentialsPolicy-as-codeSigned artifactsSBOM & model BOMCross-domain guardsIndependent red-team testing

User device boundary

Edge node boundary

Command enclave boundary

Strategic cloud boundary

Cross-domain transfer boundary

Model execution boundary

Third-party connector boundary

Administrative control-plane boundary

Non-goals for the initial product

Scope is held deliberately narrow. What VISHVAKSHA will not do in its current programme:

Authorise lethal action

Release or directly control weapons

Replace all existing C4ISR systems

Infer hostile intent automatically

Provide unrestricted autonomy to AI agents

Use consumer AI services for protected operational data

Depend on a single commercial model vendor

Centralise all classified information in one unrestricted database

Move into live operations without independent assurance and operator acceptance

Launch gates

The product cannot enter operational use until every gate below is closed.

1

Independent security review passes

2

Data provenance and classification are complete

3

AI outputs are evidence-grounded

4

High-risk workflows enforce human approvals

5

Offline operation is validated

6

Audit replay reconstructs significant decisions

7

Model rollback and drift controls are demonstrated

8

No foreign API is required for mission-critical operation

9

Software and model artifacts are signed

10

Operators are trained

11

Manual fallback is validated

12

The sponsoring organisation grants acceptance

Frequently asked questions

Can VISHVAKSHA authorise or carry out a lethal strike?

No. VISHVAKSHA is a decision-support and intelligence-fusion platform. It cannot independently authorise lethal action, directly release or control weapons, or bypass command approval. Every consequential decision remains under accountable human command.

Does VISHVAKSHA depend on foreign cloud services or APIs?

No mission-critical function requires a foreign API. MEDHA runs an indigenous, self-hosted model core alongside an approved open catalogue, inside Indian sovereign infrastructure.

What happens if VISHVAKSHA loses network connectivity?

EDGE maintains at least seven days of fully disconnected local operation, including local inference, workflow execution and encrypted evidence capture, then reconciles through conflict-aware sync once connectivity returns.

Who can approve a high-risk recommendation from VISHVAKSHA?

High-risk actions require two-person approval. Every recommendation carries its evidence, confidence and a required reviewer, and AI cannot transition a workflow into an irreversible state on its own.

Is VISHVAKSHA limited to nine use cases?

No. The nine flagship missions are where deployments start, not a ceiling. The ontology, connector model and swarm-based AI architecture are built to extend to new missions across land, air, maritime, space and cyber.

Who builds VISHVAKSHA?

VISHVAKSHA is built by EpicForge Software Private Limited, founded by Adityavardhan Nagamalli.

Talk to our governance team.

Security, accreditation and independent assurance questions welcome.

Request Briefing