KPMG: 62% of firms deploy AI agents as insurers demand oversight.

As businesses seek more complex automation, AI providers are rolling out features to make agents more capable and controllable. On 29 October, OpenAI launched a dedicated macOS application for its Codex platform, providing developers with a "command centre" to manage multiple coding agents running in parallel. The new tool allows engineers to schedule long-running AI coding tasks, automatically review code changes and track agent decisions, adding much-needed oversight to autonomous software development. This reflects a broader industry push for “multi-agent” orchestration tools that let companies supervise and coordinate AI agents tackling multi-step tasks. ([1]) For example, Microsoft’s Copilot Studio similarly emphasises governance, providing ways to group and monitor AI-driven workflows within a single interface ([2]).

These innovations expand what AI agents can do in practice. With better integration to software development platforms and enterprise systems, agents are moving beyond simple question-and-answer roles. They can now write and test code, update spreadsheets or generate presentations with minimal human input ([3]) ([4]). By building oversight features into these tools, such as permission settings and automated quality checks, vendors aim to give enterprises more confidence that agents will follow defined steps and produce reliable outputs ([5]) ([6]). For CIOs, these advances suggest that autonomous agents are becoming increasingly viable for real work, provided there are guardrails to keep their actions in check.

The growing use of autonomous AI has also drawn new scrutiny from those who insure business operations. Lloyd’s of London this year introduced standard “AI-agent” liability clauses, effectively ending what insurers called “silent AI” coverage, the inadvertent coverage of AI-related mistakes under traditional policies ([1]) ([2]). The Lloyd’s clauses require companies to maintain human-in-the-loop oversight and detailed audit trails for any high-stakes tasks handled by AI agents ([3]). Without these controls, errors caused by an autonomous system could be excluded from professional liability coverage, leaving firms exposed.

Insurers say the move is a response to a surge in AI-related incidents and uncertainty over how to price the risk of autonomous decisions. In an Insurance Journal interview last month, one cyber insurance CEO warned that “autonomous AI presents an enormous challenge for insurers” when it comes to modelling new kinds of failures ([4]). Some providers have already started adding specific AI exclusions to their standard cyber and errors-and-omissions policies, or are offering separate AI liability coverage as a new risk class ([5]). For businesses, these shifts mean that deploying AI agents without proper human supervision and governance not only increases operational risk, it could also result in higher insurance premiums or even gaps in coverage. Executives exploring advanced autonomous workflows should work closely with risk managers and insurers to ensure compliance with emerging requirements, such as robust oversight mechanisms and bias testing of AI systems ([6]) ([7]).