Shakti Yapay Zekası: Hindistanlı Kadınlar İçin Travma Bilgili Güvenlik Aracında Revizyonlar

Özet (abstract)

Version 2 (2026-08-04). Revised to the NLL Universal Paper Format v6. The original text is retained in full; nothing has been deleted. Corrections appear as marked blocks placed at the section that carries the claim, and each one states what the paper said, what the data show, the corrected claim, why it happened, and what still stands. This version corrects three claims and documents the live system's limits.The §6.1 tier escalation was not produced by the mechanism being demonstrated — the triggering word is not in the DAMP lexicon, so its cause is unknown; the system is described as evaluated when the evaluation is two prompts, with no red-teaming or clinical validation by the paper's own §7.1 and §7.5; and the BSA §65C evidentiary claim is withdrawn, since the operator controls the hash chain end to end.Verified live limits are now stated: threat detection is Latin-script substring matching, so Devanagari, Tamil, Kannada, Telugu, Bengali and Malayalam input scores 0.0, and tiers escalate without a de-escalation path. Helpline routing is correctly India-specific — Issue #1 is resolved. Everything below this line is the original description from version 1. It is retained unchanged for the record. Where it conflicts with the corrections above, the corrections stand. Two framings it repeats have since been withdrawn in full — the Bhaya Quiescence Law and the Buddhi S-Curve. Both are addressed in Maya-Meta P1, now superseded, and in the self-audit of Maya-Meta P2. Shakti-v1 is a women's safety AI companion designed for India's population of gender-based violence survivors, combining a large language model (Gemma 3 27B, fine-tuned via LoRA) with a lightweight neuromorphic spiking neural network that detects emotional and threat signals in real time across Hindi, Kannada, and English. The system's four biologically-inspired neurons — encoding fear, calm, trust, and aliveness — continuously assess incoming messages and escalate a safety tier that reshapes the AI's responses from supportive conversation through to crisis intervention, while a hash-chained immutable log preserves every exchange as tamper-evident evidence aligned with BSA Section 65C evidentiary standards. Fine-tuning on 2,000 human-authored trilingual seed samples across nine trauma-informed response categories converged in 48 minutes, with training loss falling from approximately 5.3 to 0.5 over 18 steps and zero truncated examples. This work is the second release in the Maya-Defence Series of adversarial and safety systems built on Maya neuromorphic foundations, with the Bhaya Quiescence Law confirmed in defence-grade deployment contexts. Shakti is deployed publicly at meetshakti.in with an embedded legal reference panel covering 15 Indian laws spanning the Domestic Violence Act 2005 through the Bharatiya Nyaya Sanhita 2023, making it the first neuromorphic affective AI designed specifically for Indian women's legal and emotional safety needs. Series: Part of the Maya-Defence Series — AI safety and adversarial defence systems built on Maya neuromorphic foundations. Bhaya Quiescence Law confirmed in defence-grade deployment contexts. Links: Full Series Index — venky2099.github.io Nexus Learning Labs, Bengaluru · UDYAM-KR-02-0122422 · BHASKAR IN-0526-9452JSORCID: 0000-0002-3315-7907 · VAIRAGYA_DECAY_RATE = 0.002315 — an ORCID-derived provenance mark, not an experimental parameter. Each paper's disclosure block states whether it reached a result in that paper.Canary: MayaNexusVS2026NLL_Bengaluru

Yazarların özeti; kaynağından alınmıştır. Zenodo (CERN European Organization for Nuclear Research), 2026 · DOI ↗