From drafting reports and analysing data to screening job applications, assisting customers and supporting business decisions, Artificial intelligence (AI) is becoming part of everyday professional work. But as organisations expand its use, questions around accuracy, privacy, fairness and accountability are becoming increasingly important.India’s New Delhi Declaration on AI Impact, adopted at the AI Impact Summit in February 2026, places trustworthy AI, human capital, social empowerment and international cooperation among its key priorities. The declaration states that “AI’s promise is best realized only when its benefits are shared by humanity”. It also highlights the need for secure, resilient and efficient AI systems.International frameworks are translating these broad principles into practical guidance for businesses. The Organisation for Economic Co-operation and Development (OECD) updated its AI Principles in 2024 and published its Due Diligence Guidance for Responsible AI in February 2026. The US National Institute of Standards and Technology (NIST) has also developed a risk-management framework specifically for generative AI.Together, these frameworks emphasise a common set of responsibilities: understand the purpose of an AI application, assess its risks, protect data, verify its outputs and retain human oversight.Here are 10 considerations for professionals before using AI at work.
1. Define the purpose before choosing a tool
Start by identifying the problem AI is expected to solve. Is it intended to reduce administrative work, analyse large volumes of information, improve customer service or support decision-making?Professionals should establish whether AI is appropriate for the task, whether conventional software could achieve the same result and whether its use offers a meaningful benefit. The objective should be to solve a specific problem, not adopt AI simply because the technology is available.
2. Assess the risks of each application
Not every AI application carries the same level of risk. Using a chatbot to draft an internal email is different from deploying an automated system to screen job applicants, assess creditworthiness or support medical decisions.The OECD’s due diligence guidance recommends identifying and assessing potential adverse impacts throughout the AI system’s lifecycle. Professionals should consider who could be affected, the consequences of an incorrect output and what safeguards are needed before the tool is used.
3. Protect personal and confidential information
AI tools may process information supplied by users, including customer records, employee details, personal information and commercially sensitive documents.Before using a platform, check its privacy terms, data-retention policies, access controls and whether submitted information may be used to improve its models. Confidential business information and personal records should not be entered into unapproved public AI tools.The OECD’s AI Principles identify privacy and data protection as essential elements of trustworthy AI. Organisations should ensure that information is collected and processed responsibly.
4. Check data quality and potential bias
AI systems can reproduce or amplify patterns and inequalities present in their underlying data. Incomplete or unrepresentative datasets can lead to inaccurate recommendations or unfair outcomes for particular groups.This is particularly important in recruitment, lending, healthcare and public services, where language, geography, income and social circumstances can influence results.Professionals should examine whether the data are relevant to the intended users, whether important groups are under-represented and whether the system has been tested for unfair outcomes. The OECD principles emphasise fairness, diversity and non-discrimination throughout the AI lifecycle.
5. Verify AI-generated information
Generative AI can produce fabricated references, incorrect figures, misleading explanations and convincing but inaccurate answers.NIST’s Generative Artificial Intelligence Profile describes this risk as confabulation, in which systems produce “confidently stated but erroneous or false content”.Professionals should check important claims against original documents, official records, research papers and reliable datasets. Figures should be examined for their reporting periods, units and methodology. Quotations must be verified against their original source. AI-generated material should be treated as a draft or research lead until it has been independently reviewed.
6. Retain human oversight and accountability
AI can assist with analysis and recommendations, but responsibility for consequential decisions should remain clearly assigned.The OECD principles emphasise human agency and oversight, including safeguards against unintended uses and harmful outcomes. Professionals should be able to review, question, reject or override AI-generated recommendations.A manager should not approve a hiring decision, financial assessment or customer-risk classification simply because an AI system produced it. Organisations must establish who is responsible for reviewing outputs and addressing errors.
7. Be transparent about the use of AI
People interacting with an organisation should understand when AI is involved, particularly when it generates content, influences recommendations or contributes to decisions affecting them.The OECD’s transparency principle calls for meaningful information about an AI system’s capabilities and limitations. Organisations should disclose AI involvement where appropriate, explain the system’s role and distinguish generated material from content reviewed by professionals.Transparency also means giving people a way to question or challenge decisions when AI has played a significant role.
8. Review cybersecurity and third-party risks
AI systems can introduce security vulnerabilities through their data, software, connected applications and external integrations. Tools that access company databases, emails or business systems require particular care.The UK government’s Code of Practice for the Cyber Security of AI sets out voluntary principles covering secure design, risk assessment, supply-chain security, testing and monitoring.Professionals should use approved platforms, restrict access to sensitive information and review vendor permissions. Organisations should also assess risks such as malicious prompts, unauthorised access and data leakage.
9. Consider copyright and maintain records
AI-generated content does not remove the need to consider intellectual property rights. Professionals should check whether material supplied to a model can legally be used for that purpose and whether generated content reproduces protected material.Organisations should also maintain records of the tools used, information supplied, generated outputs, approvals and human revisions. Such documentation can help establish how a report, recommendation or decision was produced and make it easier to investigate errors.The OECD’s accountability principles emphasise traceability of datasets, processes and decisions throughout an AI system’s lifecycle.
10. Monitor performance and measure the impact
Responsible AI use does not end when a tool is introduced. Changes in data, models and operating conditions can affect accuracy, security and performance.NIST’s risk-management framework encourages organisations to monitor AI systems, document incidents and review safeguards. Professionals should examine error rates, user feedback, unexpected outcomes and whether the system continues to perform as intended.They should also assess whether AI is delivering genuine productivity gains or simply shifting work, increasing costs or creating new risks. Computing requirements and environmental impacts should be considered alongside operational benefits.
From adoption to responsible use
The New Delhi Declaration, OECD principles and NIST framework approach responsible AI from different perspectives, but share an emphasis on trust, safety, human oversight and accountability. They do not establish a single universal checklist or replace national laws and sector-specific requirements.The OECD’s 2026 due diligence guidance offers businesses a structured approach: embed responsible conduct into management systems, identify and assess potential harm, prevent or mitigate adverse impacts, track results, communicate actions and provide for remediation where necessary.For professionals, responsible adoption means understanding what a system can and cannot do, protecting the information it handles, checking its outputs and retaining human control over important decisions.The measure of responsible AI is not simply whether a tool completes a task faster. It is whether the result is accurate, fair, secure and beneficial to the people affected by it.
About The Author
