Signal

Original research published this week across 107 enterprises confirmed that 54 percent have already experienced a confirmed AI agent security incident or near-miss. Only 33 percent give each agent its own scoped identity, and 70 percent of deployed agents still share credentials. A parallel study across 157 enterprises found that half have shipped an agent that passed internal evaluations and then failed a client in production. Platform announcement monitoring confirmed both datasets published simultaneously — the same week. Firms already positioning around AI governance and agent oversight are walking into client conversations with documented failure rates their clients have not yet seen.

Thesis

You have been watching your clients adopt AI tools and staying close enough to the conversation to feel useful. You know the landscape. You have opinions about which approaches work and which do not. That has been enough — because your clients have been in exploration mode, and exploration mode does not require accountability. That changed this week. Two independent enterprise studies dropped the same finding at the same time: AI agents are failing in production at scale, and most of the firms running them have no governance layer to catch it. Your clients are about to read this. Some already have. The question they are going to bring into their next conversation with a trusted advisor is not 'should we use AI' — it is 'how do we know ours is not broken.' The firms that walk in with a point of view on that question before it is asked will define the next engagement. The ones who wait to be asked are already behind.