At the HumanX event in Amsterdam, Google DeepMind’s Kareem Ayoub explained that while fully governing AI systems may be impossible, companies can establish deterministic governance perimeters around AI similar to the fixed-rule frameworks banks used to contain trading risks in the late 20th century.
- AI governance via 'fixed fence' inspired by banking risk controls
- Focus on transparency, control, and safe R&D integration
- Shift towards AI agents enables complex multi-step tasks
What happened
Kareem Ayoub, Google DeepMind’s VP of AI technical strategy, delivered insights at the HumanX conference in Amsterdam on September 23, 2026, addressing how companies can govern AI systems amid rapidly advancing capabilities. He argued that while complete governance of AI might be unattainable, organizations can implement a deterministic boundary around their AI deployments, similar to how banks in the late 1980s and early 1990s confined trading algorithms within fixed, auditable rules.
He proposed dividing the governance challenge into three key areas: defining AI's role in research and development, establishing transparent systems, and determining control over AI’s deployment. Ayoub referenced Google’s Frontier Safety Framework, which sets capability thresholds and mitigation strategies, as part of the current approach to managing AI risk. He also responded to industry discourse on pacing AI development and governance models, suggesting guardrails will require collaboration beyond the tech sector.
Why it matters
Ayoub’s perspective highlights the critical tension between the exponential growth in AI capabilities and the comparatively linear evolution of governance mechanisms. This gap, if unaddressed, could expose enterprises and society to unmitigated risks stemming from AI systems operating without robust oversight.
The analogy to banking regulation underscores the necessity for creating controlled environments where AI can safely be tested and utilized. Establishing such boundaries enables businesses to take advantage of AI’s powerful potential—such as accelerating drug discovery or translating ancient texts—while maintaining accountability and transparency. It also signals a shift in AI productization from simple models to sophisticated agents that integrate tools and internet access to complete complex workflows.
What to watch next
Stakeholders should monitor the evolution of industry-led regulatory frameworks similar to FINRA in finance, aiming to provide standardized oversight for AI developments. DeepMind’s ongoing initiatives, including the recently launched DeepMind Institute and the Gemini program, will also be key indicators of how AI governance and innovation coalesce.
At the organizational scale, the best practices identified by Ayoub—rapid development of safe sandboxes, alignment with model improvements, comprehensive measurement of outcomes, and leadership engagement with AI tools—will be critical signals of responsible AI adoption. Additionally, how companies manage morale and talent amid AI’s rapid evolution, especially as leadership roles shift and startups emerge from veteran AI researchers, will shape the AI governance landscape.