Risks in Multi-domain AI Assistants: A Literature-Based Analysis with Focus on the Indian Context

Multi-domain artificial intelligence (AI) assistants—systems capable of handling diverse tasks spanning conversation, code generation, information retrieval, planning, and tool use—have rapidly become integral to personal, enterprise, and public-sector workflows. Their versatility expands the attack surface and amplifies risks that single-purpose models do not fully exhibit. This literature-based research paper synthesizes findings from more than twenty peer-reviewed studies, industry frameworks, and policy documents published primarily between 2021 and 2026. It examines technical risks (prompt injection, excessive agency, hallucination, data leakage), socio-technical risks (bias, misinformation, representational harms), and systemic risks (supply-chain vulnerabilities, multi-agent failure modes). Special attention is given to the Indian context, including the implications of the Digital Personal Data Protection (DPDP) Act, 2023, linguistic and cultural diversity challenges, digital public infrastructure exposure, and emerging governance gaps. A comparative review of 22 key works is presented. The paper concludes with recommendations for layered technical controls, organizational practices, and policy interventions suited to India's regulatory and socio-economic landscape.