Mitigating against four types of GenAI risk  

The four types  

1. Misuse 

– The unethical or illegal exploitation of GenAI capabilities for harmful purposes such as scams and misinformation campaigns. 

2. Misapplication

– Where GenAI prioritizes plausibility of accuracy and creates inaccurate outputs – known as “hallucination”. It becomes an issue when users improperly depend on it or misapply GenAI tools. 

3. Misrepresentation

– Where GenAI output created by a third party is purposefully used and disseminated, despite doubts about its credibility or authenticity. 

4. Misadventure

– When content is consumed and shared by users who are not aware of its inauthenticity, such as the sharing of deepfake content. 

Risk mitigation 

Each of the above risks presents unique challenges, yet there are actions that leaders can take to protect their organizations against them, bearing in mind the different types. 

1. Ensure alignment between organizational values and AI principles 

Establish clear guidelines and principles – such as transparency, fairness, accountability, and safety – for GenAI use to ensure it does not cause any personal or societal harm. 

2. Mandate all entities that create GenAI content to watermark their output 

If content is generated by an AI system, it should be clearly watermarked as such so users can distinguish between AI and human-created outputs. 

3. Create a controlled GenAI environment within the organization 

This can be done by creating your own fine-tuned GenAI tools, such as an LLM, and adding extra layers of privacy management tools to the architecture to ensure no personally identifiable information finds its way into the system. 

4. Provide GenAI demystification and awareness-training opportunities for all 

Provide training programs to raise awareness on the safe and responsible consumption of GenAI content, coupled with internal guardrails around its use and policies on when it can and cannot be used.  

5. Validate AI output through labeling and warning mechanisms 

In addition to clearly watermarking all AI-generated content, put other mechanisms in place to cross-check and verify content, and ensure these mechanisms are sufficiently robust to detect both hallucinations and deliberate misapplications of GenAI content. 

6. Set up damage-mitigation plans for situations that are not contained 

Despite your best efforts, not all threats will be contained, and some problems might “go public”.  In such instances, it is important to have an internal task force in place that can quickly understand, prioritize, and control risks, and communicate with all relevant stakeholders.  

Further reading 

A real leader’s guide to AI 

GenAI: the future belongs to those who pause and reflect 

Navigating GenAI’s ethical risks to score competitive value  

AI’s five strategic tensions and how to resolve them 

The right AI for the job: Generative vs. legacy – when to use each  

Why boring is beautiful with GenAI