DBS's Former CTO: 'AI Scaled Risk More Than It Created New Categories of Risk'

Forrester analyst Frederic Giron published a piece built around a conversation with David Walker, former CTO of Westpac and DBS (Development Bank of Singapore), making a specific governance argument: AI models increasingly supply the intelligence behind workflows, but humans still determine where AI gets applied and how its value gets measured. Walker’s core claim — “AI scaled risk more than it created entirely new categories of risk” — is an argument for folding AI risk assessment into existing governance frameworks rather than standing up parallel AI-specific processes.

Walker also flags a scaling problem specific to human oversight: reviewers grow complacent when AI performs correctly repeatedly, weakening manual controls at exactly the point automated controls become essential to catch the failure. DBS is cited as an example of spreading AI-transformation skill organization-wide rather than centralizing it in a specialist team — framing itself, in Walker’s words, as “a 22,000-person startup.” The piece pushes for agent-based work redesign — giving agents outcome objectives instead of teaching them to mimic existing processes — citing a data-migration workflow cut from six weeks to four days using that approach.

The oversight-complacency problem Walker names is the same one Jason Lemkin’s SaaStr data implies from a different direction. In an August analysis, Lemkin found most CEOs claiming to “rebuild for AI” have shipped features and an agent or two without touching core product, pricing, or sales motion — a 40%-of-the-way-there gap that looks a lot like governance and org design lagging behind deployment. McKinsey’s research on AI transformations and trust points at the same mechanism from the employee side: roughly one in five employees report anxiety about AI-driven change, rising to one in four among middle managers, and workers with low trust in organizational support are 1.5 times more likely to report that anxiety.

For revenue and enablement leaders, Walker’s reframe is the useful one to steal: don’t build a separate AI governance function. Fold agent oversight into whatever risk process already catches a human analyst’s mistakes — because the failure mode is the same complacency, just scaled.