When Google contributes to the AI Governance debate, I listen
Google published a White Paper on A Pragmatic Approach to AI Governance in America (June 2026). Google argues for a differentiated governance model based on risk and capability. The most notable proposal is the creation of a federally supervised Frontier AI Regulatory Organization (FARO), an industry-funded body that would set safety standards, define capability-based benchmarks and oversee audits of the most advanced AI models, particularly where cyber security and CBRN risks arise.
Rather than relying on training compute thresholds alone, Google proposes capability-based assessments and progressively more substantive audits.
For AI systems below the frontier threshold, the paper takes a different position: apply and, where necessary, adapt existing laws rather than create entirely new AI-specific regulatory regimes. Google repeatedly emphasises regulating harmful outputs rather than the underlying science or development process.
From a privacy perspective, the most controversial element may be the proposal to move from “Privacy by Design” towards “Privacy by Innovation”. The paper advocates stronger use of Privacy Enhancing Technologies, contextual integrity and privacy controls embedded in agentic AI ecosystems.
However, this raises important questions for organisations operating under the GDPR. Privacy by Design is a legal obligation under Article 25 GDPR. Replacing mandatory requirements with market-driven incentives reflects a distinctly US regulatory philosophy and sits uneasily with the European understanding of data protection as a fundamental right. The paper also says little about the legal basis for collecting and processing training data, a question that remains central to many ongoing AI governance discussions.
Overall, the paper presents a pragmatic and technically informed framework. But whether it achieves the right balance between innovation, security and the protection of individuals will ultimately depend on how such proposals are implemented in practice, and how they interact with existing regimes such as the EU AI Act and the GDPR.