Responsibilities
The standards we hold ourselves to
Building AI for the defence of the realm is a position of trust. These are the ethics and policies that govern how we work — not aspirations, but commitments we design within from the first line of code, and against which you may hold us. They sit alongside our Safety notice, which sets out how the software itself is built to be used safely.
01
Lawful use only
We build for lawful, authorised missions and decline anything outside them. Our tools are for the institutions charged with the security of the realm and its people — never for unlawful surveillance, for use against citizens' rights, or for any purpose beyond a legitimate mandate.
02
Human accountability
A qualified, accountable human stays in command of every consequential action. Our software recognises, classifies and recommends; it does not authorise, engage or act on its own. There is no autonomous use of force and no machine authority over decisions that affect life or safety.
03
Data protection & custody
Sensitive data belongs to the institution that owns it. We design to keep it there — on your own hardware, at the edge, offline if it must — and to collect the minimum needed. We publish, in plain terms, where each datum goes and where it stays, and we honour your evidence-chain and data-protection protocol.
04
Honest representation
We describe what is real. Capabilities are products and services in development, not claims of fielded systems; benchmark figures are reported with their conditions and known limits, never as guarantees. Where a system is early, narrow or unproven in your environment, we say so — before you depend on it.
05
Privacy & proportionality
Every inference is a computer-generated lead requiring human corroboration, not a finding of fact. We design against fishing and dragnet misuse, and our analysis tools are honest about their limits — structure is not intent, co-location is not a meeting. Use must be proportionate to a legitimate, authorised purpose.
06
Fairness & non-discrimination
We work to avoid systems that disadvantage people by who they are. Because our models are deterministic and auditable, their behaviour can be examined and challenged. We treat bias and disparate impact as defects to be found and fixed, not risks to be hidden.
07
Dual-use & export responsibility
Defence and security software is inherently dual-use, and we treat that as a duty, not a footnote. We exercise diligence over who we work with and the end-use of what we build, and we supply with applicable export-control and trade-compliance obligations in mind. As a sovereign capability, what is fielded is controlled by the nation that fields it.
08
Security & responsible disclosure
We secure what we build and welcome responsible disclosure. If you find a vulnerability, a misuse, or behaviour that falls short of what we describe, tell us — we read every report, take it seriously, and respond. Please do not include classified or operationally sensitive material in an email.
Oversight & accountability
Designed to be questioned
A standard you cannot inspect is a slogan. Ours are built to be checked.
Because our models are deterministic and closed-form, a machine call can be traced back to the evidence behind it — a glass box, not a black box. Operators retain the ability to override, pause or shut the software down. Deployment follows the customer's own evaluation, accreditation and acceptance, under the lawful chain of command — never our assurance alone.
Raise a concern
If something here isn't being honoured, tell us.
Misuse, unsafe behaviour, a capability falling short of what we describe, or a vulnerability — we want to hear it, and we will respond. Responsible disclosure is welcome.
Commitments, held by design
For lawful defence and security use only. Deployment is subject to applicable export-control regulations and an operator-controlled human-in-the-loop. Capabilities shown are products and services in development — not claims of fielded systems. Abhedya is a venture of Aitele Research LLP · Bihar, India · DPIIT-recognised startup. Nothing on this page is a warranty, a guarantee of performance, or legal advice. Read together with our Safety notice, Legal notice and Terms.
Last updated: June 2026