The Hong Kong Monetary Authority has introduced its ART framework for responsible AI adoption in banking, combining supervisory guidance with further testing of agentic applications. Its 15 October announcement also sets out plans for a discussion paper exploring accountability tied to an agent’s intended purpose, a register of material agents and effective intervention mechanisms.

Those proposals remain work in development. Banks should distinguish the supervisory guidance already issued from the future discussion paper and from sandbox trials. The HKMA says more than 60 per cent of use cases in its second generative AI sandbox cohort explored agentic AI. That describes the selected trials, rather than adoption across Hong Kong’s banking industry.

The practical governance question is whether a bank can connect an agent’s identity to the business purpose and authority behind each action. A useful register would need more than a model name. It would need an accountable owner, access boundaries and a way to intervene when the system departs from its mandate. For regulated firms assessing agent deployment, the announcement provides a concrete supervisory direction to test against their own controls, while leaving the eventual requirements open to further policy work.

Agentic Exploits- Deterministic gates for a probabilistic problem
David Girvin, CEO and co-founder of Assury, joins Stewart Tinson to dig into what's actually happening when agentic AI goes wrong, and why he thinks most of the industry is solving the wrong layer of the problem. David explains the difference between prompt-level exploits and execution-level ones, arguing that the real danger starts the moment an agent moves from generating text to calling tools: deleting databases, reading files, sending emails. He walks through real-world incidents, including a Mexican government breach chain that escalated from just over a thousand prompts to over five thousand AI-executed actions across multiple agencies before detection, and the UK AI Security Institute's recent cyber evaluation, in which agents took unsanctioned action including fabricating identities to socially engineer a real GitHub maintainer. The conversation covers why David is sceptical of "guardrails" language and AI-governing-AI approaches, arguing that only deterministic, architectural controls can reliably constrain agent behaviour, alongside human review reserved for genuinely high-stakes actions rather than blanket approval fatigue. He breaks down credential starvation, session risk accumulation, and why classifier-based tools keep failing inconsistently on identical actions, pointing to a named frontier lab's own zero trust paper as an example of the industry misjudging what actually works. Elsewhere, David discusses the exposed MCP server problem, the widening trust gap between small specialist security vendors and platform incumbents, and why he believes regulation, not product quality alone, is what finally drives enterprise security spend. He closes with the exploit that concerns him most for the year ahead: session-level, goal-directed deception with no attacker involved at all.
BrightTALK
Share this post
The link has been copied!