A Dharmic Ethics Checklist for Building Better Technology

When you’re deciding whether to ship an AI feature, buy a platform, or add another layer of automation, the easiest question is also the weakest: can it work? A Dharmic review asks harder questions before capability becomes habit. What kind of action does the system make easier? Whose freedom expands? Whose burden stays hidden? Can the harm be repaired?

This is not a case for rejecting technology. It is a practical method for keeping the maker responsible for the made. You can use it in a product review, procurement decision, institutional policy, or personal technology choice.

Key takeaways

• Judge a technology by the agency it preserves, not only the capability it adds.
• Treat prosperity and enjoyment as legitimate aims, but require both to remain governed by Dharma and directed toward freedom rather than compulsion.
• Translate Ahimsa, Satya, Asteya, Brahmacharya, and Aparigraha into testable requirements, release evidence, and stop conditions.
• Trace first-, second-, and third-order consequences across physical, psychological, social, ecological, and spiritual life.
• When uncertainty is high, prefer smaller scope, reversibility, repairability, and meaningful human control.
• Do not approve a consequential system without an accountable owner, a way to challenge decisions, a remedy for harm, and a credible exit or shutdown path.

Begin with agency, not capability

Technology does not become unethical only when it malfunctions. It can work exactly as designed and still weaken attention, conceal responsibility, concentrate power, or make dependence feel like convenience. The first task is therefore not to list features. It is to define the form of human agency the system must protect.

The Katha Upanishad gives you a useful diagnostic image. The senses are horses, the mind holds the reins, and discerning intelligence acts as the charioteer. Applied to technology, more speed and stimulation do not amount to progress if the reins have passed from the user to the system. A recommendation engine, notification layer, or automated workflow should assist deliberate action. It should not quietly become the author of that action.

Before approving a project, complete this sentence in plain language: “This technology helps a defined person perform a defined action without creating a hidden dependency or transferring an unreasonable burden to someone else.” If your team cannot name the person, the action, the dependency, and the burden, the proposal is not yet ethically clear enough to approve.

The four Purusharthas sharpen that test. Artha includes material capacity and prosperity. Kama includes enjoyment, satisfaction, and legitimate desire. Neither is rejected. Both must operate within Dharma, while Moksha supplies the longer horizon: freedom from fear, delusion, domination, and compulsion. A Dharmic technology ethic therefore asks four connected questions:

• Artha: What real capacity, livelihood, efficiency, or material benefit does this create?
• Kama: What legitimate pleasure, ease, connection, or satisfaction does it enable?
• Dharma: What duties, limits, relationships, and forms of truthfulness must govern those benefits?
• Moksha: Does repeated use leave the person more capable of deliberate action, or more dependent on prompts, rewards, surveillance, and fear?

This prevents a common category error. Engagement, growth, convenience, and revenue may demonstrate that a system is effective at producing a response. They do not establish that the response is worth producing. If a product succeeds by making refusal difficult, confusing consent, or stimulating compulsive return, commercial success does not cure the ethical defect.

Read the design through the three Gunas

Rajas, Tamas, and Sattva can be used as design diagnostics rather than labels placed on people. Every complex system may contain all three.

• Rajas reveals acceleration. Look for speed, novelty, competition, stimulation, and pressure to act. Ask where the system pushes a user faster than reflection can follow.
• Tamas reveals obscurity. Look for hidden costs, confusing controls, institutional inertia, buried consequences, and responsibility that disappears between departments.
• Sattva reveals clarity. Look for truthful feedback, understandable choices, proportionate pacing, visible consequences, and conditions in which a person can decide without manipulation.

The practical aim is not to eliminate energy or ambition. It is to govern Rajasic speed so that it does not leave Tamasic harm behind. For every feature that accelerates action, add a corresponding clarity control: a natural stopping point, a review step for an irreversible choice, an explanation of why a recommendation appeared, or an uncomplicated way to refuse.

Translate Dharmic principles into release requirements

An ethical value has little force until it changes a default, permission, metric, contract, test, or release decision. “We value trust” is an aspiration. “A user can decline secondary data use without losing the core service” is a requirement that can be inspected.

The Yogic Yamas and Niyamas provide a disciplined way to turn moral language into engineering and governance obligations.

Use the table as a release worksheet. For every applicable discipline, write the obligation, identify the mechanism that could violate it, name the evidence needed, and assign the person who can block release. If nobody has blocking authority, the obligation is ceremonial.

Inspect the craving loop, not just the interface

Buddhist dependent origination directs attention to conditions and feedback loops. In a digital product, sensation, desire, repeated action, and habitual formation can reinforce one another. The ethical question is not merely whether a button is attractive. It is whether the entire sequence turns a passing impulse into a durable compulsion.

Map the loop from trigger to repeated behaviour. Mark where the system predicts a vulnerable response, intensifies craving or aversion, removes a natural stopping point, and rewards return. Then introduce an interruption: reduce unnecessary prompts, stop autoplay or endless continuation, offer a genuine pause, or let the person choose a non-ranked view. Right intention concerns the purpose of the design; right livelihood concerns whether the organisation’s prosperity depends on increasing suffering or confusion.

Require the strongest dissenting view

Jain Anekantavada treats truth as many-sided. For a technology review, this means no metric, model, department, or social group should be mistaken for the whole. A high average benefit can conceal severe harm borne by a small group. A technically accurate model can still rest on a narrow definition of success. A compliant consent screen can still be practically unintelligible.

Do not ask stakeholders only whether they agree. Ask each group to state what the others are failing to see. Record the strongest unresolved objection in the decision document, along with the evidence that would confirm or weaken it. Include affected non-users, frontline workers, communities carrying environmental costs, and people who must appeal the system’s decisions. Dissent is useful design information, not an obstacle to be managed away.

Test whether service reaches everyone who bears the cost

Sikh teachings of Simran and Seva keep technological power tied to remembrance and service. Sarbat da bhala – the well-being of all – is a demanding distribution test. It asks whether the people who bear a system’s risks also share its benefits, and whether those with the least institutional power receive a meaningful voice and remedy.

Add two columns to every business case: “Who receives the gain?” and “Who carries the cost?” Name groups rather than writing “society” or “users.” If one group receives convenience while another absorbs surveillance, displaced responsibility, dangerous work, pollution, or exclusion, redesign the arrangement before calling it service.

Run a Karmic Impact Assessment before you commit

Karma is not a vague promise that good intentions will produce good outcomes. In a technology decision, it directs attention to action, consequence, habit, and the conditions each action creates for the next one. A Karmic Impact Assessment makes that chain visible while architecture, incentives, contracts, and defaults can still be changed.

Run the assessment before a major commitment and repeat it before release. Revisit it when the purpose, data, model, affected population, scale, or operating environment changes. It complements technical safety, security, legal, and domain review; it does not replace them.

1. State the intended good. Describe the human capacity or burden you intend to improve. Do not substitute a delivery metric such as adoption, automation, or engagement for the actual good.
2. Name everyone affected. Include users, non-users, workers, families, institutions, communities, ecosystems, and people who may inherit a dependency or waste stream later.
3. Trace the orders of consequence. First-order effects follow directly from use. Second-order effects arise when people and institutions adapt their behaviour. Third-order effects appear when the technology changes incentives, power, infrastructure, or what society treats as normal.
4. Inspect every domain of harm. Examine physical safety, psychological compulsion or fear, social trust and power, ecological extraction and waste, and spiritual effects on truthfulness, dignity, attention, and freedom.
5. Separate benefit from distribution. Record who receives each gain, who bears each risk, whether they consented, and whether they can realistically refuse.
6. Test uncertainty and tail risk. Identify assumptions with weak evidence and rare outcomes whose damage would be catastrophic. Do not hide either inside an average score.
7. Design for reversal and repair. Specify how the system can be rolled back, how a person can exit, how an automated decision can be challenged, how damage will be repaired, and who pays for that repair.
8. Write the decision conditions. Choose whether to proceed, narrow the scope, add safeguards, conduct a limited pilot, redesign, or stop. Attach an owner, monitoring signal, review trigger, and shutdown authority to the decision.

Do not collapse unlike harms into one score

A single numerical ethics score creates false precision. A modest convenience benefit cannot automatically cancel serious deception, irreversible exclusion, or catastrophic risk. Keep the impact domains separate. Show the reasoning and the disagreement. Decision-makers should be able to see which duty is being traded away, who authorised the trade, and why it was considered acceptable.

Use explicit Dharma gates instead. A gate is a condition that must be satisfied regardless of projected upside. Suitable gates include no deceptive consent, no undisclosed secondary use of personal data, no engineered compulsion as the core revenue mechanism, no consequential automated decision without accountable oversight, and no deployment where an affected person lacks a practical route to correction or remedy.

Where knowledge is incomplete, choose the option that preserves future choice. A limited scope is easier to reverse than universal deployment. A repairable device is less binding than a sealed one. A system that collects only what it needs creates less future exposure than one that accumulates information merely because storage is available. Restraint is not hostility to innovation; it is the discipline that keeps experimentation from becoming an irreversible burden.

Make Dharma enforceable in high-risk systems

The same framework applies across technologies, but different systems conceal harm in different places. Your release gate should follow the mechanism of risk rather than rely on a generic ethics statement.

Attention platforms: return the reins to the user

Attention systems can convert craving and aversion into predictable engagement. Outrage, novelty, sexual stimulation, social comparison, and anxiety become especially dangerous when the design removes stopping cues and continually selects the next stimulus.

• Identify every mechanism whose purpose is to prolong use after the person’s stated task is complete.
• Replace endless continuation with natural stopping points and an explicit choice to continue.
• Expose why content is being recommended and provide a usable alternative to behavioural ranking.
• Do not disguise advertisements, urgency, refusal paths, or the consequences of sharing data.
• Measure whether people accomplish their intended task, not only how long they remain available for monetisation.
• Give users controls that remain effective after updates rather than quietly returning them to the most extractive default.

Digital Pratyahara offers a parallel personal discipline: periodically withdraw from engineered stimulation so attention can again become available for deliberate use. For a product team, however, a user’s spiritual discipline must never become an excuse for manipulative design. The maker remains responsible for the conditions the maker creates.

Artificial intelligence: connect every output to accountability

AI alignment is moral as well as technical because the system’s objective, training assumptions, deployment context, and institutional incentives determine what “good performance” means. A model can optimise its assigned measure while obscuring harm outside that measure.

For consequential AI, require bias auditing, adversarial safety testing, accountable human oversight, and interpretability proportionate to the stakes. Model cards and system cards can describe capabilities and limitations. The Karmic Impact Assessment must go further by tracing who is nudged, selected, hired, denied, watched, or silenced, and by exposing the assumptions used to justify those actions.

• Name the human who owns the decision produced or supported by the system. “The model decided” is not an acceptable accountability structure.
• Define what that person can inspect, override, pause, and correct. A nominal human in the loop has little value if the interface, workload, or policy makes disagreement impractical.
• Give affected people understandable notice when AI materially shapes a consequential outcome.
• Create a challenge route that reaches someone empowered to change the result and repair harm.
• Specify the failure signal that narrows or stops deployment. Monitoring without a decision threshold merely observes damage.
• Reject deployment where nobody can explain the relevant basis of a high-stakes outcome well enough to assume responsibility for it.

Cybersecurity: protect without normalising unnecessary aggression

Defensive security serves Ahimsa by protecting life, privacy, trust, and critical infrastructure. Least-privilege architecture, privacy by default, and responsible disclosure turn that duty into operating practice. Offensive escalation requires a stricter necessity test because avoidable retaliation, indiscriminate effects, and expanding surveillance can reproduce the harm security was meant to prevent.

Ask whether each permission is necessary, whether sensitive access expires, whether one compromised account can reach unrelated systems, and whether incident response preserves civil liberties. Secrecy may protect a vulnerability during remediation, but it must not become a permanent shelter for negligence or unaccountable power.

Hardware and infrastructure: account for the material afterlife

Panchamahabhuta consciousness refuses to treat space, air, fire, water, and earth as an invisible warehouse for technological growth. A device is not immaterial because its interface feels frictionless. Materials are extracted, components are assembled, energy is consumed, and discarded equipment remains somewhere after the user stops seeing it.

• Require replaceable components where practical and document how the product can be disassembled.
• Ask how long repair information and replacement parts will remain available.
• Prefer open standards that prevent useful equipment from becoming waste through avoidable lock-in.
• Demand toxin transparency and an end-of-life reclamation route before procurement.
• Assign producer responsibility for collection, repair, reuse, or safe disposal instead of transferring every cost to households and local communities.
• Include ecological damage and future waste in the product decision, not only in a sustainability statement issued after launch.

Governance: give ethical objections somewhere to go

Multi-perspectival deliberation has little force unless an institution can act on what it learns. Sabhas and Sanghas offer a useful orientation toward hearing, testing, and reconciling different views. Modern technology governance needs corresponding structures: multi-stakeholder oversight, transparent standards, independent audit where the stakes warrant it, meaningful consent, and grievance redressal.

Before release, verify that the governance record contains all of the following:

• A clear purpose and a list of uses that remain prohibited even if they become profitable.
• A named executive, technical, and operational owner for foreseeable harms.
• Representation from people affected by the system, including those who do not purchase or directly use it.
• The strongest dissenting assessment and a written response to it.
• Evidence supporting claims about safety, efficacy, privacy, and benefit.
• A consent or notice process that an ordinary affected person can understand and act upon.
• A complaint channel connected to investigation, correction, and repair.
• Monitoring that covers harm and distribution, not only uptime, adoption, and revenue.
• A review trigger for material changes in data, model, purpose, scale, or affected population.
• A person with authority to narrow, suspend, or end the system when Dharma gates fail.

Bring a one-page version of this framework to your next technology decision. Write down the intended good, every affected group, the three orders of consequence, the non-negotiable duties, the evidence for release, the remedy for harm, and the condition that will stop deployment. If the page fills with slogans rather than owners and controls, pause. The right next step may be a narrower scope, less data, more friction, stronger human judgement, a repairable design, or no deployment at all.

That pause is not a failure of innovation. It is the moment intelligence takes the reins.

References

• DharmaRenaissance Blog – When Inventions Rule Their Makers: Dharmic Ethics to Reclaim Agency in a Tech Age