In partnership with

Have questions about securing AI? You're not alone.

Machine identities are multiplying, and cyber threats continue to evolve. That's why IAM leaders are heading to Gartner Identity & Access Management Summit 2026, December 7-9 in Las Vegas.

Join 1,700+ peers, hear directly from Gartner analysts, and discover practical strategies for governing AI agents, advancing PAM and IGA, utilizing identity visibility and intelligence platforms (IVIPs) and preparing your IAM organization for the future.

Register with code IAM21P2 and save $400 on the standard rate.

Explore Conference

-THE WIRE  THIS WEEK IN IDENTITY

N°01 · MARKET

Proofpoint announced an agreement to acquire Varonis for $6.3 billion, framing the deal as connecting people risk with data access. Read that as a bet that identity, AI security, and data governance belong on the same platform. A year ago those were three separate budget lines and three separate vendors. The market is already deciding this question. The organizations waiting for a governance framework before they act may find the framework is built into a platform they did not buy.

→ Proofpoint

N°02 · MARKET

Workday signed a definitive agreement to acquire C1, the company that used to be ConductorOne. Nothing has closed, terms were undisclosed, and Workday expects to close by the end of January. The system of record for joiners, movers and leavers just bought the governance layer for agents, betting that every agent registration ends up looking like a new hire event. The announcement also states the execution gap outright, from the company writing the check. It cites C1’s own survey of 508 IT and security leaders, 95% already running agents autonomously, then sets Deloitte’s finding that 21% have a mature model for governing autonomous agents next to it. Two unrelated surveys, and the distance between those numbers is the story of Year One.

→ Workday

Lots to dig into this week, also a big update for you, so stay to the end! Quick shoutout to Gartner for being sponsor this last couple of episodes, it’s been so awesome to work with them, makes sure to click the link if you’re looking to save some bucks on the Gartner IAM conference this year. Ok, let’s get to the good stuff!

Twenty issues ago, in early June, I opened this newsletter with a question that felt dramatic at the time: what happens to identity governance when the workforce is no longer made up of people? Half of the organizations the Cloud Security Alliance had surveyed reported no clear ownership of their AI identities. Not weak ownership. None. A different 47% reported NHIs with credentials untouched for over a year. The agents were running. Nobody had the list.

That was Year One, week one. Twenty issues later, the question is different. The question is where the field actually moved and where it only looked like it moved.

Here is what actually changed.

The inventory problem is more tractable than it was in June. Not solved. Tractable. Twelve months ago, most shops were manually walking IdP exports and interviewing platform teams, which meant the list was wrong by the following Wednesday. A handful of vendors can now enumerate agent identities across cloud and SaaS surfaces without requiring a human to walk them. The critical caveat is that discovery is not governance. You can have a beautiful, auto-refreshing inventory and still have zero accountability structure on top of it. The teams that moved most in Year One used the tooling to see and then used the Principal Owner model to assign. The teams that bought the discovery tool and stopped there know more about their exposure. Managing it is a different problem.

The credential migration is further along than the conversation suggests. The shift from long-lived API keys to short-lived workload identity is the clearest technical win of the year, because it is a binary you can measure: either your agents authenticate with a credential that expires in hours or they authenticate with one that expires never. The enterprise security community treated SPIFFE/SVID and AWS IAM Roles Anywhere as aspirational during the RSAC wave of 2026. By Q3, the practitioners in this community who ran a pilot were using them in production. For the teams that made the shift, the long-lived shared key is behind them.

The audit trail is the area where vendor positioning and practitioner reality are furthest apart. Every platform that launched an "Agentic IGA" capability this year included "full session audit" in the pitch deck. What ships is something narrower: gateway-level logs, maybe tool-call records if you run their native integrations, nothing that correlates across the five planes an agent actually touches. The organizations with real forensic capability built most of it themselves, using the correlation layer architecture we outlined in Issue 16, because the integrated product does not exist at a maturity level worth buying. The gap between what a platform claims on the procurement call and what you can actually do at 2am is still wide.

The ownership model is where the maturity distribution in this community is most bimodal. Some readers have a working Principal Owner architecture: named humans, documented attestation cadence, an escalation path that goes somewhere when the owner leaves the organization. Others are still in the phase where ownership means whoever originally provisioned the agent. The teams in the first group went through an uncomfortable conversation with their platform owners about accountability, and they had to win it without a regulatory mandate behind them. That changes now. EU AI Act Article 15 enforcement is live as of August 2, and the traceability requirements are going to force the ownership conversation into organizations that have been avoiding it.

What is still theater: certifying agents like people. This is the most honest thing I can say at the end of twenty issues, and it comes with a correction to how I have framed it all year. Almost nobody is running a quarterly access review on their agents today. The certification campaigns that exist were built for employees, and agents sit outside their scope. The theater is therefore in the plan rather than in anything already running. The campaign machinery is already bought, the auditor already recognizes it, and extending it to cover agents is the path of least resistance, so that is the design most programs will drift into by default. It will hit the same wall Issue 18 described, because ninety days of agent activity is a behavioral archive and not a document an owner can read. I made the case there for event-triggered certification and I stand by it. What I would add now is that the window to choose it is wide open precisely because so little has been built yet. The number in this issue’s own Wire says as much. Four out of five organizations already running AI in daily production have no mature model for governing autonomous agents, and that is not a field quietly completing quarterly reviews. If the 2027 plan is to extend the employee campaign to cover agents, the plan is to produce evidence for an auditor instead of signal for your team.

Where this goes in 2027: the conversation moves from governance architecture to governance operations. Year One was about establishing the vocabulary and running the first pilots. Year Two is about running the program at scale, integrating agent identity events into the SOC workflow without drowning the SOC in noise, and surviving the first real regulatory exam. The organizations that used Year One to build the foundation will have a running start.

Year One closed the awareness gap. Every serious practitioner in this community now understands that agents are identities, that IGA was not built for them, and that the governance model has to be rebuilt not patched. What Year One did not close is the execution gap: most programs have a working inventory and an ownership model on paper, and nothing yet built to certify what their agents can reach. The credential migration is the clearest win. The audit trail is the clearest remaining gap. And agent certification is the piece with the most room left to get right, because almost none of it has been built. That is the honest state of the field on October 21, 2026.

Twenty issues is a long time. I am glad we spent it here. The field needed the diagnosis, it needed the architecture, and it needed the practitioner-level specificity that turns a governance framework into something a team of three can actually run. What I do not know yet is where you are landing as Year Two starts: which piece of the program is hardest to sustain, not to start, and what is the thing that keeps slipping even when the intent is solid? Hit reply and tell me. I read every one.

P.S ( Kinda) As I promised, a quick update and an ask. So I entered a contest to win an IHeart Radio produced podcast series. As some of you know, seven years ago I started my media company with the vision to create digital media in different platforms, and tell the stories that weren’t being told. This is a chance to take that to the next level. Currently i’m in 1st place ( as of me writing this), but voting ends tomorrow for this round, and I’d appreciate your vote. ( They count double today!!). You can vote for free, here’s the link: https://tophost.org/2026/david-lee-2.

See ya next week.