Abstract
During elections, referendums and emergencies, recommendation, ranking, search and advertising algorithms decide what millions of people see. A quiet mid-election tweak, a swapped model, a hidden experiment on voters, or a feed changed after review can shift public opinion. Today such manipulation is mostly discovered afterwards, through audits and reports, when the harm is already done.
Based on Indian Patent Filing: 202631127058, filed 25 October 2026 (specification Part 2), building on the Mothership application PCT/IB2026/055615, filed 4 June 2026 and publicly disclosed 16 July 2026.
This disclosure is designed to prevent algorithmic manipulation before content reaches people. It works in three steps:
- Measure what actually ran. A protected component measures which algorithm configuration actually produced each feed, ranking or recommendation. It turns that into a fingerprint without revealing the platform’s proprietary model.
- Inspector check. During a protected period declared by an authorised election or emergency authority, an inspector compares that fingerprint with the approved configuration. If it matches, the inspector issues a signed receipt.
- Gatekeeper check. At the point of delivery, a gatekeeper rebuilds the exact output about to be sent and releases it only with both the receipt and a separate one-time key.
These checks stop:
- secret mid-election changes;
- unapproved models;
- hidden experiments;
- side-door delivery routes;
- fake or replayed mode switches;
- last-moment edits.
People’s own posts are never subject to approval. Only algorithmic boosting is checked, and disputed decisions can be re-examined through counter-verification.
Regulatory relevance (AI-generated). The approach could provide per-output, machine-verifiable evidence for:
- EU AI Act: election-influencing AI as high-risk, record-keeping and human oversight.
- Digital Services Act: election risks, recommender transparency, audits and researcher access.
- Digital Markets Act: fair ranking and audited profiling.
- Other EU rules: the Political Advertising Regulation, GDPR and the European Media Freedom Act.
- India: the IT Rules as amended in 2026 and Election Commission of India advisories.
- Elsewhere: U.S. state election laws, the UK Online Safety Act and China’s algorithm-filing rules.
Disclaimer. The invention is the work of its human inventor. This title, abstract and the accompanying summary, including all legal and regulatory references, were prepared with AI assistance and have not been independently verified. Readers must cross-check every statement against the original filing and the official legal texts, and reach their own final conclusions. Nothing here is legal advice or a statement of compliance. The application is pending. Licence: CC BY-NC 4.0. Contact: Sangam Das, [email protected]. The inventor is open to acquihire or strategic collaboration.
Keywords: AI safety; algorithmic manipulation prevention; election integrity; democratic processes; recommender systems; ranking transparency; algorithm fingerprint; runtime attestation; change control; drift detection; hidden experiments; output verification; freedom of expression; content neutrality; counter-verification; EU AI Act; Digital Services Act; Digital Markets Act; political advertising regulation; GDPR; European Media Freedom Act; IT Rules 2026; Election Commission of India; Online Safety Act; algorithm filing
Creative Commons License

This work is licensed under a Creative Commons Attribution-Noncommercial 4.0 License
Recommended Citation
Das, Sangam Kumar, "AI Safety for Democratic Elections: Preventing Algorithmic Manipulation by Verifying, Output by Output, That Platforms’ Recommendation and Ranking Algorithms Are the Approved Ones — Without Judging Speech", Technical Disclosure Commons, ()
https://www.tdcommons.org/dpubs_series/13022