February 2024

Penalties of the EU AI Act: The High Cost of Non-Compliance

The EU has proposed the Harmonised Rules on Artificial Intelligence (EU AI Act) to lead the world in AI regulation and build trust in AI systems. The Act sets out a risk-based approach for AI systems, defines three levels of risk, and subjects certain AI systems to transparency obligations. The Act also introduces a three-tiered model of penalties for violators, with the heftiest fines imposed on those who violate the prohibition of specific AI systems. The EU AI Act has gone through an extended consultation process and has been subject to amendments throughout. Penalties can be issued to providers, deployers, importers, distributors, and notified bodies. The EU AI Act emphasizes proportionality and offers lower penalties for SMEs and startups. There is no union-wide central authority for imposing fines on AI operators, so penalties depend on the national legal system of Member States. The fines for providers of GPAI models and Union bodies are imposed by the Commission and the European Data Protection Supervisor, respectively.

Operationalising Safety in Generative AI: Model Evaluations and Algorithm Audits

Ensuring the integrity, safety, security, and reliability of generative AI models is crucial for organizations developing and deploying them. Two important processes to achieve this are model evaluations and algorithm audits. While model evaluations assess a model's efficacy across various parameters, such as performance levels and risks, algorithm audits involve independent third-party assessments of reliability, risk detection, and regulatory compliance. Both processes should be used jointly to build the evidence base of a model's safety and risk mitigation capabilities. As regulatory momentum to legislate generative models accelerates, companies must proactively ensure they fulfill their obligations. Holistic AI's LLM Auditing product is a solution that can help identify and address issues such as blocking serious risks, detecting hallucinations and stereotypes, preventing offensive language and toxicity, and providing readability scores.

FTC Launches Inquiry into Generative AI Investments: What you need to know

The Federal Trade Commission (FTC) has issued orders to Alphabet, Amazon, Anthropic, Microsoft, and OpenAI to provide information about their investments and partnerships in generative AI companies, citing concerns about how these investments may distort innovation and undermine fair competition. The companies must provide information on agreements and related documents, interaction and influence, analyses and reports, documents related to exclusivity and access, materials provided to government entities, specifications for document production, use of technology, and contact information and communication. The FTC aims to better understand the competitive landscape and potential implications of AI collaborations to ensure fair competition and prevent practices that could stifle innovation.

Making Sense of China’s AI Regulations

China's AI market is worth $23.196 billion USD in 2021 and is expected to triple to $61.855 billion by 2025. The government expects AI to create $154.638 billion USD in annual revenue by 2030. China has been introducing AI regulations since 2021, with three distinct regulatory measures enforced at the national, regional, and local levels. These regulations aim to regulate the proliferation of AI and its innovative use cases. They cover a wide scope, including deepfake technology, internet information service algorithmic recommendation management, and generative AI. The regulations seek to mitigate potential harms associated with AI, and they set a crucial precedent for other jurisdictions to follow.

Approaching the tape: The EU’s final strides in the AI legislation marathon

The EU AI Act reached a provisional agreement on 9 December 2023 and was unanimously endorsed by Coreper I on 2 February 2024, making it likely to be official once voted on by the European Parliament in April 2024. After adoption, there will be a two-year grace period for implementation and enforcement, during which the Commission will conduct the AI Pact to encourage early commitment to the Act's rules and principles. Companies should begin preparing for compliance with the Act to maximize alignment. Holistic AI offers governance, risk, and compliance platforms and innovative solutions to help companies navigate the Act's rules and requirements.