AI Suite | AIoniser

⚖️

Case 1 · Legal Technology

SmartAttorney AI — Legal Compliance Assistant

A multi-agent AI system providing legal guidance, regulatory compliance support, and jurisdiction-aware advice accessible by individuals, SMEs, civil society organisations, and public institutions across any legal system. Deployable as a natural-language interface in any language or jurisdiction worldwide.

⚠ Assessing Harms Potential negative social impacts of AI-generated legal guidance

Dimension	Individual	Collective	Societal
Type of Harm	● High Over-reliance on AI legal advice without professional verification, leading to adverse outcomes in disputes, filings, or contracts Misapplication of jurisdiction-specific rules — AI may conflate legal systems, causing incorrect compliance decisions Denial or misunderstanding of individual rights through inaccurate guidance on labour, property, or criminal matters Gradual erosion of personal legal literacy as users substitute AI outputs for genuine understanding of the law	● Medium Groups adopting standardised but incorrect AI-generated templates amplify errors across entire communities or organisations Marginalised populations — including women, informal workers, and linguistic minorities — disproportionately harmed when AI is biased toward dominant legal frameworks Collective erosion of trust in digital public services if AI-generated guidance visibly leads to legal failures	● High Inconsistent application of legal rules by AI undermines public understanding of the rule of law Displacement of qualified legal professionals over time, reducing access to nuanced human advocacy Entrenched inequalities if the tool remains inaccessible to populations without reliable internet or devices Risk that dominant-language training data inadvertently overrides local customary or indigenous law
Source of Harm	Hallucination of legal citations — AI fabricates case references, statutes, or regulatory articles that do not exist Training data lag — model knowledge does not reflect recent legislative amendments or court rulings Translation and interpretation errors for users querying in languages other than the model's primary training language	Uniform advice generation producing collective legal errors across similarly-situated groups at scale Absence of cultural, customary, or indigenous law sensitivity in AI trained predominantly on formal legal corpora No community-level feedback mechanism to detect and correct systematic biases in AI outputs	Regulatory gaps in many jurisdictions — no established framework governing AI-generated legal advice Limited accountability of commercial LLM providers under local consumer or professional liability law Concentration of access in urban or high-income contexts, widening existing justice gaps Absence of cross-jurisdictional AI auditing standards for legal technology
Measure of Harm	Rate of AI responses producing incorrect legal guidance as measured against verified legal benchmarks Frequency of user-reported adverse outcomes following AI legal advice (complaint tracking) Hallucination rate per 100 queries in domain-specific legal accuracy evaluations	Volume of incorrectly-filed regulatory documents or contracts attributable to AI template errors Proportion of underrepresented communities reporting disproportionate adverse outcomes Digital trust index among target user groups measured through periodic surveys	Reduction in formally-qualified legal professionals serving low-income populations over time Number of legislative, judicial, or regulatory challenges related to AI-generated legal documents Access gap metric: percentage of target population unable to use the tool due to connectivity or device barriers
Mitigation of Harm	Mandatory prominent disclaimer on every response: "AI-generated guidance — always verify with a qualified attorney before acting" Source citation requirement: every response must reference the specific statute, regulation, or legal principle invoked Daily query limits to discourage sole reliance and encourage professional consultation Multi-language support with validated translations to reduce interpretation-induced errors	Referral pathways to accredited legal aid organisations, bar associations, or ombudsman services for complex queries Periodic bulk-output audits by independent legal professionals to detect systematic error patterns Co-design testing with diverse user groups — including informal-economy workers, women, and rural communities — before deployment	Proactive engagement with national bar associations, ministries of justice, and regulatory bodies to develop AI legal advice governance frameworks Publication of model cards, training data provenance, and quarterly bias audit reports under open-access policy Offline or low-bandwidth access modes to extend reach to underserved populations Human-in-the-loop escalation for high-stakes query categories (property rights, criminal matters, family law)

✦ Amplifying Benefits Positive social value generated by AI-powered legal access

Dimension	Individual	Collective	Societal
Type of Benefit	● High Value Democratised access to legal information for individuals who cannot afford professional legal services Empowerment of entrepreneurs, employees, and tenants to understand and assert their rights Reduced time, cost, and anxiety for citizens navigating unfamiliar or complex legal processes Multi-language support enables speakers of minority or regional languages to access legal guidance in their own language	● High Value Strengthened SME and civil society ecosystem through accessible, consistent compliance guidance Community organisations better equipped to manage contractual and regulatory obligations Populations in geographically remote or underserved areas gain equivalent access to legal knowledge available in major cities	● High Value Contribution to national digital justice and access-to-law policy goals across any jurisdiction Potential reduction in court backlog as citizens resolve minor disputes or clarify rights without litigation Improved regulatory compliance and investor confidence through consistent, accessible guidance Positions deploying governments and organisations as leaders in responsible AI for public services
Source of Benefit	24/7 availability with zero marginal cost per query, removing time and cost barriers to legal information Natural language interface lowering literacy and technical barriers compared to formal legal documentation Personalised, context-aware responses that adapt to the user's stated situation and jurisdiction	Multi-jurisdictional knowledge base enabling cross-border compliance support for international businesses Scalable architecture allowing simultaneous service to large user populations without proportional cost increases Multi-agent CrewAI pipeline enabling cross-checking of retrieved legal information before response generation	Alignment with global access-to-justice frameworks including UN SDG 16 (Peace, Justice, Strong Institutions) Aggregated, anonymised query data provides governments with real-time insight into legal needs and regulatory gaps Open-source deployment model enables government adaptation, localisation, and extension
Measure of Benefit	Number of users successfully completing a compliance action or resolving a legal question following AI guidance User-reported legal confidence score before and after interaction (self-assessed legal literacy) Average time saved per user compared to seeking traditional legal consultation	Number of SMEs and organisations achieving successful regulatory compliance attributable to AI guidance Geographic distribution of active users as a proxy for equity of access Query volume from non-dominant languages as a proxy for inclusion of linguistic minorities	Measurable reduction in minor civil cases in jurisdictions where the platform is actively deployed Adoption of platform query insights by law reform bodies or ministries to prioritise legislative updates Recognition by international legal technology or access-to-justice organisations as a responsible AI model
Amplification of Benefit	Integration with CV and career advisory AI tools to provide employment law and contract guidance to job seekers globally Session memory enabling multi-turn legal consultations for complex, evolving legal needs PDF export of AI-generated legal summaries for use in official correspondence or submissions	Partnerships with legal aid NGOs, bar associations, and regulators to embed AI guidance in public onboarding workflows Paralegal and community advocate training programmes using SmartAttorney as a shared reference tool Integration with financial advisory AI (e.g. Bank CRM) to support businesses needing combined legal and financial compliance guidance	Scalable to national public service deployments through government procurement and open-source licensing Cross-border harmonisation of AI legal guidance standards through regional trade bodies and international legal organisations Academic and policy research partnerships to continuously improve accuracy, fairness, and legal coverage Anonymised query pattern data contributed to law reform bodies to surface systemic gaps in public legal knowledge

🏥

Case 2 · Clinical Health Technology

AI Clinical Guidelines Assistant

A RAG-powered multi-agent AI system providing clinicians, health workers, and patients with accurate, verifiable guidance drawn from authoritative clinical guidelines applicable to any disease area or health system. Deployable for ART, communicable disease, chronic condition, maternal health, or any protocol-driven clinical domain worldwide.

⚠ Assessing Harms Potential negative social impacts of AI-assisted clinical guidance

Dimension	Individual	Collective	Societal
Type of Harm	● Critical Risk Incorrect treatment protocol guidance leading to clinical failure, adverse drug reactions, or patient harm Patient self-medication or premature treatment changes based on AI responses without clinical validation Breach of patient confidentiality if query content reveals sensitive health status to third parties or through data compromise Stigma amplification if AI responses frame sensitive conditions (infectious disease, mental health, reproductive health) in discriminatory or stigmatising language	● Critical Risk Health worker over-reliance leading to atrophy of clinical reasoning skills over time, particularly in resource-limited settings Community health workers applying AI guidance beyond their authorised competency scope, causing collective clinical errors Differential access between well-resourced and under-resourced facilities deepening inequities in care quality	● Critical Risk Erosion of clinician authority and public trust in national health programmes if AI guidance contradicts official protocols Risk of undermining hard-won health outcome gains if AI-introduced variance reduces adherence to evidence-based treatment standards Absence of validated AI clinical tool frameworks under most national health regulatory bodies as of 2026 Potential misuse of aggregated health query data for surveillance, profiling, or insurance discrimination
Source of Harm	RAG retrieval failure returning incorrect document chunks, producing contextually wrong clinical answers Guideline version lag — AI knowledge base not updated to reflect the most recent clinical protocol revisions Model hallucination producing plausible but fabricated drug dosages, contraindications, or interaction warnings	Reliance on a single or limited knowledge source creating blind spots where local adaptation or updated international guidelines are not reflected Absence of weight-based or age-adjusted dosing calculators in MVP versions, increasing paediatric and geriatric dosing risk Language barriers when clinical staff query in a language not well-represented in the model's training data	Most jurisdictions lack a regulatory pathway for classifying and approving AI clinical decision support tools Processing of sensitive health queries through third-party LLM APIs may violate national health data protection requirements Lack of interoperability with national health information management systems limiting clinical audit and oversight
Measure of Harm	Clinical accuracy rate per query type against gold-standard clinical benchmark assessments by domain specialists Rate of AI responses flagged by clinical reviewers as potentially harmful per 100 queries (adverse signal rate) Number of reported patient safety incidents plausibly attributable to AI guidance errors	Proportion of facilities using AI clinical tools without a mandatory clinical supervisor review layer Access differential in AI guidance accuracy and availability between well-resourced and under-resourced facilities Clinician self-reported reduction in diagnostic confidence over time (dependency index)	Trends in key national health programme KPIs in facilities where AI is deployed versus control facilities Number of regulatory alerts, suspensions, or interventions issued by health authorities regarding AI clinical tools Volume of data protection complaints or breaches linked to AI clinical query processing
Mitigation of Harm	Mandatory clinical disclaimer on every response: "Always apply clinical judgement. Consult official guidelines directly before any prescribing or treatment decision." Source-grounded RAG — every response must cite the specific section, page, and version of the guideline document it draws from Daily query limits to prevent dependency and reinforce the tool's role as a reference, not a replacement for clinical judgment Automated guideline version monitoring with knowledge base updates aligned to official guideline publication cycles	Mandatory clinical supervisor review layer for all community health worker and lay-provider deployments Pre-deployment training module clearly distinguishing AI as a reference tool versus a diagnostic or prescribing system Multi-language interface with clinically validated translations to remove language-based access barriers Weight-based and age-adjusted dosing calculation modules included in all clinical deployment versions	Proactive engagement with national health ministries, medical councils, and digital health regulatory bodies for tool classification and approval Data processing agreements with LLM providers specifying health data residency, retention, and non-use for training Quarterly independent clinical audit of AI response accuracy by authority-appointed clinical reviewers API integration with national health information systems for anonymised query-pattern monitoring and clinical oversight

✦ Amplifying Benefits Positive social value of AI-assisted clinical decision support

Dimension	Individual	Collective	Societal
Type of Benefit	● High Value Clinicians gain instant, verified access to protocol guidance without manually navigating lengthy guideline documents Nurses, midwives, and community health workers in remote facilities supported with protocol knowledge beyond their standard training scope Patients access plain-language explanations of their treatment plans, improving adherence and informed consent Reduced cognitive burden on health workers managing complex multi-morbidity cases under high-workload conditions	● High Value Standardised protocol application across all facility types reduces geographic and resource-based variation in care quality Community health workers equipped to deliver accurate first-line counselling aligned with current clinical standards Under-staffed facilities in remote or low-resource settings gain 24/7 decision support equivalent to specialist consultation	● High Value Contribution to national and global health targets (e.g. WHO Universal Health Coverage, disease-specific elimination goals) through improved protocol adherence Reduced rates of treatment failure lower downstream costs of second- and third-line interventions across health systems Positions deploying health systems as global leaders in responsible AI for clinical decision support Open, replicable model applicable to any disease area or clinical guideline set worldwide
Source of Benefit	High-fidelity vector index of authoritative clinical guidelines enabling precise, citation-backed retrieval Agentic document parsing ensuring accurate extraction of clinical tables, dosing charts, and decision algorithms Two-stage RAG pipeline (context creation + search prompt generation) maximising clinical answer accuracy	Scalable architecture enabling concurrent access by large numbers of health workers without infrastructure bottlenecks Real-time SSE streaming providing immediate responses for time-sensitive clinical queries Cloud-neutral deployment model enabling access from any device with internet connectivity globally	Institutional partnership model with authoritative bodies (WHO, national health ministries, professional colleges) as knowledge source owners Alignment with global digital health strategies including WHO Global Strategy on Digital Health 2020–2025 Academic publication potential — documenting RAG-based clinical AI methodology applicable to any health system
Measure of Benefit	Clinical accuracy score per query category versus specialist clinician review benchmarks Time-to-answer for clinical queries: AI (under 10 seconds) versus manual guideline search (5–15 minutes) Health worker satisfaction and perceived usefulness scores measured through periodic surveys	Reduction in inter-facility variation in protocol adherence before and after AI introduction Geographic reach: proportion of target facilities or health workers with active access to the platform Training completion rates and post-training competency scores for AI-assisted clinical protocols	Trends in key clinical outcome indicators in facilities where AI decision support is deployed versus control groups Number of other health systems or countries adopting the RAG-clinical AI model (replication and scaling index) Formal endorsement or adoption by WHO, regional health bodies, or national health authorities
Amplification of Benefit	Integration with patient-facing health companion applications for personalised treatment adherence support AI-assisted patient education on treatment plans, side effects, and follow-up schedules in the patient's own language Extension to additional disease areas and specialties using the same RAG infrastructure	Embedding in community health worker training and referral workflows at scale Integration with national health information management systems to contextualise guidance with facility-level epidemiological data Peer-learning networks of health workers sharing AI use cases and best practices across facilities	Scaling to national public health system deployment through ministry of health procurement and open-source licensing Replication of the knowledge base model across WHO AFRO, EMRO, SEARO, and other regional office guideline repositories Pre-commercial procurement and public innovation funding pathways to accelerate responsible deployment globally Peer-reviewed publication establishing the model as a reference architecture for AI clinical decision support in resource-limited settings

Social Impact Assessment Matrix Report

SmartAttorney AI — Legal Compliance Assistant

AI Clinical Guidelines Assistant