AI that serves institutions without undermining the values they protect.
Merdot builds AI across a wide range · from an everyday assistant to institutional intelligence infrastructure. These principles define how we build it, how we constrain it, and where we draw lines that hold regardless of what a client requests.
Core principles
Seven commitments that govern how we build and deploy AI.
They apply across the ecosystem · Merdot AI, Merdot Track and MIOS Pinaka.
Human decision authority
Our systems produce intelligence · they do not make consequential decisions. Every classification, forecast and recommended response is advisory. Human analysts and decision-makers retain full authority over the actions that follow. Merdot is built to augment human judgment, never to replace it.
Transparency
Outputs carry confidence scores, evidence chains and source attribution. Analysts can inspect the signals behind a classification, the network evidence behind an attribution and the data behind a forecast. No black-box output reaches a decision-maker unexplained.
Fairness
Systems are designed to detect coordinated inauthentic behaviour · not to suppress legitimate speech, journalism or civil-society activity. We do not configure our platforms to target individuals or groups on the basis of religion, ethnicity, caste, political affiliation or lawful advocacy.
Privacy
We collect and process only the data necessary for a stated function, and we work at the network and narrative level rather than individual surveillance. Personal data is not retained beyond operational necessity. We build to comply with applicable data-protection law, including India's DPDP Act.
Security
Intelligence is sensitive by nature, so security is treated as a first-order requirement · access controls, encryption in transit and at rest, tenant isolation and least-privilege engineering. We aim to hold ourselves to recognised security standards as the platform matures.
Accountability
Consequential outputs are logged with full evidence chains, and those logs are designed to resist tampering. We take responsibility for how our systems behave, and institutional clients accept responsibility for their use of the platform within their legal authority.
Lawful use
Our products are built for detection, analysis and defence · not offensive information operations. We reserve the right to decline or terminate any engagement that violates these principles, applicable law, or that causes demonstrable public harm.
Safety architecture
Safety built into the engine, not bolted on.
Some constraints operate at the model level and are designed so they cannot be switched off by configuration or client instruction.
Confidence thresholds
The analyst will not promote uncertain findings to false certainty. Low-evidence outputs are surfaced as uncertain, not dressed up as high-confidence conclusions.
Attribution constraints
Individual-person attribution requires multi-signal confirmation. The analyst will not assign individual-level blame from a single signal.
No demographic profiling
Systems cannot be configured to produce intelligence products that profile individuals by religion, ethnicity, caste or political affiliation.
No offensive content generation
The analyst does not generate disinformation, synthetic social posts or content designed for narrative manipulation.
Audit trails
Consequential outputs are logged with their full evidence chain, and those logs are designed so client administrators cannot quietly delete them.
Use policy
What our platforms will and will not be used for.
Permitted use
- Detection and analysis of coordinated inauthentic behaviour and influence operations
- Monitoring of information environments for crisis situational awareness
- Election-integrity monitoring and disinformation detection by authorised oversight bodies
- National-security narrative threat monitoring by authorised government agencies
- Enterprise reputation defence and corporate risk intelligence
- Academic and policy research on information operations, under appropriate agreements
Prohibited use
- Generating, amplifying or deploying disinformation or synthetic content
- Surveillance targeting based on religion, ethnicity, caste or political belief
- Suppression or targeting of journalists, civil society or legitimate opposition
- Offensive information operations against domestic populations
- Electoral manipulation on behalf of any political party
- Any use that violates applicable national or international law
Questions about responsible use?
We are committed to being clear about how our systems work and what they will not do. If you have questions about these principles or a specific deployment, reach out directly.