Navigating New Global AI Laws and Their Impact on Risk and Safety Assessment

The Rise of AI Regulations

AI is now an integral part of various industries, including being used to enhance risk and safety assessments. From predictive models for risk analysis to automation in safety protocols, AI's capability to analyze large volumes of data swiftly and with precision is reshaping these fields. However, as AI becomes more embedded in decision-making processes, governments are stepping up their efforts to regulate its use to prevent biases, protect privacy, and ensure safety.

Some key areas of focus in upcoming AI regulations include:

1. Transparency – Companies will be required to explain how AI systems make decisions, ensuring that their processes are transparent and accountable.
2. Data Privacy – AI systems handle large volumes of sensitive data, so regulations will focus on how this data is collected, stored, and used, ensuring that it aligns with privacy standards.
3. Bias and Fairness – AI models can sometimes perpetuate or amplify biases. New regulations will require businesses to address potential biases in their AI models and ensure they provide equitable outcomes.
4. Safety and Risk – Given the stakes in fields like safety assessment, AI systems must be rigorously tested to ensure they don't pose new risks to human health or the environment.

How AI Regulations Could Affect Risk and Safety Assessments

Risk and safety assessments are a critical part of industries like cosmetics, pharmaceuticals, chemicals, and manufacturing. However, with new global AI laws, businesses must navigate additional compliance requirements, particularly regarding how AI is used in decision-making for risk evaluation and safety protocols.

Here are a few specific impacts to consider:

1. Data Handling and Transparency
As AI increasingly guides decisions about safety and risk assessments, the need for transparency in how these systems operate will become critical. If your business uses AI to evaluate risks (e.g., in chemical exposure assessments or environmental risk management), you will need to document and explain how your AI models arrive at decisions. Companies will likely be required to share information about their data sources, model training, and the assumptions underlying their AI systems.

Ex. The pharmaceutical industry uses AI to predict drug safety profiles, such as potential side effects. Big pharma companies employ machine learning algorithms to analyze vast datasets, including clinical trial data, to predict adverse reactions. Under new AI regulations, these companies will be required to provide full transparency on the models used, including how data is selected and how the models predict outcomes, ensuring the public and regulators can scrutinize and trust the results.

2. Bias in Risk Models
AI systems used in risk assessments must be carefully designed to avoid perpetuating biases that could lead to unsafe or discriminatory decisions. For instance, if AI models used in environmental risk assessments rely on historical data that’s incomplete or skewed, the results could fail to account for certain populations or regions. As AI regulations tighten, businesses will need to ensure their models are not only accurate but also fair and unbiased.

Ex. In environmental risk assessments, AI models are used to predict the impact of pollution on various communities. A model trained primarily on data from urban areas might miss nuances in rural or indigenous communities. This could result in policies that don't fully protect vulnerable populations. To comply with emerging AI regulations, companies would need to re-evaluate their models, ensuring they are trained on diverse datasets and mitigate any biases related to geography, income, or race.

3. Regulatory Compliance and Validation
The upcoming laws will require businesses to ensure their AI models comply with regulatory standards. This will involve validating the algorithms used for risk assessments, ensuring they align with established safety protocols, and verifying that the AI doesn’t compromise human oversight. In practice, this means that businesses must be ready to regularly audit and update their AI systems to ensure continued compliance.

Ex. The cosmetics industry, particularly larger companies increasingly relying on AI for the safety testing of new products. AI models simulate skin irritation or toxicity without animal testing, which helps companies comply with evolving regulations like the EU's ban on animal testing. However, under new AI regulations, companies must ensure that these models are validated against real-world data and that their outputs align with existing safety standards like the International Nomenclature of Cosmetic Ingredients (INCI). Regular audits may be necessary to demonstrate that these models remain compliant.

4. Human-in-the-Loop Oversight
Many global AI laws are introducing requirements for "human-in-the-loop" (HITL) systems, meaning AI decisions must be reviewed and validated by humans, particularly in high-stakes areas like safety. Businesses will need to balance AI’s efficiency with human judgment to ensure that safety decisions are made with full context and consideration.

Ex. In the chemical manufacturing industry, AI models are used to predict the toxicity of new substances based on molecular structures. While AI can rapidly assess hundreds of compounds, the final decision about safety must still be made by a trained toxicologist. Under new AI regulations, companies will need to have a system where human experts are involved in the final review of any AI-generated risk assessments, ensuring that these decisions are not purely automated.

Practical Steps for Businesses to Prepare

With new AI laws coming into play, businesses in risk and safety assessments need to take proactive steps to stay ahead of the curve. Here are some practical actions to consider:

1. Audit AI Systems Regularly
   Regularly review and audit your AI models to ensure they comply with emerging regulations. This includes ensuring transparency in decision-making, assessing data quality, and validating model outputs. Keeping records of these audits will be essential for demonstrating compliance.
2. Ensure Data Integrity and Privacy
   Review your data collection, storage, and processing practices to ensure they comply with global data privacy regulations, such as the GDPR in Europe or similar frameworks in other regions. This includes ensuring informed consent from data sources and protecting sensitive data.
3. Address Bias and Fairness
   Conduct thorough testing to identify and mitigate any biases in your AI models. This may involve diversifying training datasets, rethinking the features used in models, and introducing fairness metrics into your assessments. Proactively addressing bias will help ensure your risk assessments are reliable and equitable.
4. Implement HITL Processes
   Develop processes that ensure human oversight in critical AI decisions. For instance, even if an AI model provides a risk assessment, a qualified safety officer or risk assessor should review and validate the outcome before any decisions are made.
5. Stay Informed About Regulations
   AI laws are rapidly evolving, so it's crucial to stay informed about developments in the regulatory landscape. Regularly consult with legal experts and regulatory bodies to ensure your systems are always in compliance with the latest rules.

Specific AI Laws and Regulations to Watch

Here are some specific AI laws and frameworks currently in development or already making an impact globally:

1. European Union - Artificial Intelligence Act (AI Act)
   The EU is pioneering comprehensive AI regulation with the AI Act, which came into effect in April 2021. It classifies AI systems into risk categories, requiring companies to comply with strict transparency and accountability measures for high-risk AI systems, which include those used in safety-critical areas. This will impact industries like pharmaceuticals and manufacturing, where AI is increasingly used in risk assessments.
2. United States - Algorithmic Accountability Act
   Introduced in 2019, this proposed law would require companies to audit their AI systems for potential biases and fairness issues. If passed, this law would have significant implications for industries like healthcare and financial services, where AI plays a central role in risk assessment.
3. China - AI Governance Principles (2022)
   China has laid out guidelines for the ethical use of AI, with a strong emphasis on ensuring that AI systems are safe, fair, and transparent. This framework impacts industries operating in China, particularly tech and consumer goods companies that use AI to manage safety and risk in their products.
4. United Kingdom - The National AI Strategy
   The UK’s approach includes promoting safe AI development and setting guidelines for AI safety, especially in risk-critical industries. The UK aims to balance fostering innovation with strong regulatory frameworks to manage potential AI-related risks.
5. OECD Principles on Artificial Intelligence (2021)
   The Organisation for Economic Co-operation and Development (OECD) has provided recommendations for the responsible development of AI, encouraging transparency, fairness, and accountability. These principles are influential globally and serve as a reference for many national AI regulations.

Final Thoughts

As AI regulations continue to evolve, businesses must stay informed and proactive to ensure their AI systems, especially in risk and safety assessments, comply with new laws. A critical component of this is ensuring high-quality data management, which underpins AI's transparency, fairness, and reliability. Implementing systems that support data integrity, version control, and seamless integration—such as SaferWorldbyDesign’s Software Tools—can help businesses harmonize and share data while maintaining compliance. These tools ensure that risk assessments are based on trusted data, enabling reproducibility and traceability to meet regulatory standards. By taking action now to improve data management and ensure transparency, businesses can build AI systems that not only comply with evolving regulations but also enhance trust and accountability in their decision-making processes.