Okta gathered the industry in Las Vegas this week and unveiled a "Blueprint for the Secure Agentic Enterprise," complete with a "Blueprint Alliance" of CrowdStrike, Zscaler, AWS, Salesforce and Google. It is a genuinely useful set of ideas wrapped in the most crowded word in security. And it arrived in the same fortnight that Cyera raised another $400M — its third round of 2026 — at a valuation that, notably, did not move. Two very different signals about where this market's money and language are going, and both worth reading closely.
So this issue covers the last two weeks in full: Oktane and why we're skeptical of alliance theater, Cyera's remarkable cash intake and the operational-efficiency question it raises, and a vendor deep-dive on One Identity — a good product inside an ownership structure we think is about to change hands. Plus a piece on the quiet mistake underneath the agent-governance gold rush: everyone is racing to govern non-human identities while the human identity program that makes it all work sits half-built. We refresh the Buzzword Index too. Let's get into it.
two weeks, distilled — & why it lands on your desk
At Oktane (Sep 22–24), Okta announced its Blueprint framework built around three questions — where are my agents, what can they connect to, what can they do — plus "Okta for AI Agents" to discover known and "shadow" agents, register them as first-class identities in Universal Directory, and "instantly revoke access… the ultimate kill switch." It extended its Identity Security Fabric (SSO, governance, provisioning) to agents, proposed "Cross-App Access" (XAA) as an open MCP extension, and closed its Permiso acquisition. The headline organizational move: a "Blueprint Alliance" with CrowdStrike, Zscaler, AWS, Salesforce and Google.
On Sep 22, Cyera announced a $400M Series G extension led by Goldman Sachs Alternatives. It is the company's third raise this year: a $300M Series F in January (~$9B), a $540M Series G in June (~$12B), and now this extension — at the same ~$12B valuation as June. That's roughly $1.2B raised in 2026 and more than $2B in total for a company founded in 2021, now ~1,500 employees across 18 countries with six acquisitions behind it (including Oasis Security for ~$1B).
Bloomberg reported that Clearlake Capital — which owns Quest Software, and with it One Identity — is exploring options that include combining One Identity with RSA's SecurID into a new company and selling a stake to another private-equity firm. One Identity has passed through Dell, then Francisco Partners, then Clearlake (2021, ~$5.4B for Quest), took a $350M Clearlake debt-refinancing infusion in 2025, and installed a new CEO, Praerit Garg, the same year.
On Sep 24, Omada — our Issue 15 spotlight and current #3 on the Vendor Index — announced it had acquired EmpowerID. Terms were not disclosed. The official framing is strategic and forward-looking: Omada gets runtime authorization for AI agents (an AuthZEN-compliant policy decision point, an MCP gateway, 300+ connectors) and "a credible path" into PBAM, PAM and Zero Trust. Notably, the announcement said almost nothing about EmpowerID's own IGA suite or its existing customers — one reference deployment governs 569,000 identities — and analysts immediately flagged the risk that the EmpowerID product line gets absorbed and its heavily-customized deployments orphaned. Remember what EmpowerID was: a KuppingerCole "Overall Leader" in its IDaaS-IGA Leadership Compass. That is a steep arc — from named leader to an acquisition that's pitched entirely around one piece of the tech and silent on the rest.
Beneath the framework branding, the interoperability work is genuine. Okta's Cross-App Access extends MCP with audience-bound tokens; it dovetails with CrowdStrike's SPIFFE/CAEP posture at Fal.Con and the July MCP authorization spec. For all the marketing sameness, the plumbing under agent identity is standardizing — which is the thing buyers should actually reward.
The market's whole vocabulary this fortnight was non-human: agents, NHIs, workload identity, machine credentials. Meanwhile the incidents that keep landing — including the Hugging Face token exposure we revisit in section 04 — trace back to ordinary human-identity hygiene: over-scoped tokens, no lifecycle, no real identity program behind the machine accounts. The governance you're being sold for agents only works if the human layer under it is already sound.
Our Buzzword Index refresh (below) reads a higher composite this fortnight, driven almost entirely by Oktane. "Fabric," "first-class identities," "shadow agents," "kill switch," "control plane," "blueprint" — Okta's launch language is now nearly indistinguishable from SailPoint's, Saviynt's and CrowdStrike's. Cross-vendor claim overlap on the core agentic themes nudged to about 93%.
Oktane's Blueprint Alliance, and why we grade alliances on delivery, not press releases
Okta's "Blueprint Alliance" — CrowdStrike, Zscaler, AWS, Salesforce, Google — is the latest in a genre the identity market has run for years: the big-tent security alliance, announced with a keynote and a shared slide, promising an integrated future. We are skeptical, and the skepticism is earned. The overwhelming majority of these alliances have not turned into a single thing a buyer can deploy differently the next morning.
Okta's Blueprint is a coherent framework: treat agents as first-class identities, discover the shadow ones, govern them through the existing Identity Security Fabric, and keep a fast revoke path. The technical pieces — Okta for AI Agents, XAA on MCP, the Universal Directory agent category — are real product. The Alliance, by contrast, is a statement of intent among six companies that already integrate with each other in the ordinary course of business. Even Okta's own leadership signaled restraint: co-founder and CEO Todd McKinnon described joint go-to-market and shared economic incentives as "premature." When the company announcing the alliance is publicly tapping the brakes on what it actually means commercially, a buyer should take the cue.
The identity and security industry has produced a long parade of alliances, consortia and "ecosystems" over the past several years. A few — mostly the standards bodies — produced durable, useful specifications. Most of the commercial alliances produced a launch moment and little a customer could point to a year later. The pattern is consistent enough to treat as a base rate:
| What gets announced | The promise | What usually materializes |
|---|---|---|
| A branded alliance of 4–6 large vendors | "Deep, seamless integration across the stack" | The same API-level integrations that already existed, now with a logo lockup |
| Joint go-to-market language | "One motion, better outcomes for shared customers" | Rarely a joint SKU, joint support, or joint accountability; sales teams still compete |
| A shared "framework" or "blueprint" | "An open, vendor-neutral reference architecture" | A reference architecture that centers the announcing vendor's platform |
| A standards commitment inside the alliance | "We'll converge on open protocols" | This part often does deliver — but it delivers through the standards body, not the alliance branding |
Note the last row, because it's where we're fair to Okta. The genuinely valuable output of this announcement — Cross-App Access on MCP — is real precisely because it's headed for an open specification, not because it's inside a branded alliance. The plumbing standardizes; the alliance mostly markets.
This isn't bad faith; it's incentives. A commercial alliance among large, independently public (or PE-owned) vendors has no shared P&L, no shared roadmap authority, and no mechanism to force a partner to prioritize a joint integration over its own quarter. Each member optimizes for its own pipeline. The alliance has a marketing budget and a logo; it does not have the one thing that produces integrated products, which is a single team that owns the outcome and can be fired for missing it. That is why the durable interoperability in this industry comes from standards (SAML, SCIM, OIDC, SSF/CAEP, and now MCP) and from acquisitions — one company buying another and merging the codebases — and almost never from an alliance.
Take the framework seriously and the alliance lightly. The three Blueprint questions are good governance questions; ask them of every agentic vendor, Okta included. But do not let "member of the Blueprint Alliance" appear on a vendor's scorecard as a capability. It isn't one. If Okta and CrowdStrike genuinely ship a jointly-supported, jointly-accountable integration with a name, a SKU and a support path, reward that when it exists — and hold the receipt from this week to check against it a year from now. We will.
Cyera raised a third time in 2026 — where is all that money going?
Cyera is one of the most impressive growth stories in security — real ARR, real logos, a category it helped define. It is also raising money at a pace that deserves a clear-eyed question no keynote will ask: for a data-security platform that must continuously scan, classify and monitor its customers' data at cloud scale, what do the unit economics actually look like — and is this much capital a sign of momentum, or of burn?
Three rounds in nine months is unusual even by late-stage standards. What stands out is not that Cyera is raising — great companies raise into strength — but the frequency and the flat mark on the most recent one.
Here is the part the funding headlines skip. Cyera's core value proposition — a "data security platform" that discovers, classifies and continuously monitors sensitive data across a customer's clouds, SaaS and on-prem — is, architecturally, one of the more compute- and storage-intensive things you can sell in security. To do it well you have to:
• Scan at volume, continuously. Classifying petabytes of customer data, and re-scanning as it changes, is sustained compute and egress — not a one-time index. Every large customer is an ongoing cloud bill, not just a license.
• Run AI/ML classification. The "AI-powered" classification that differentiates modern DSPM is inference at scale, which is GPU/accelerator cost that rises with usage, not with headcount.
• Store and process metadata. The data map itself — lineage, sensitivity, access — is a large, hot dataset that has to be kept current per customer.
• Integrate broadly. Coverage across every cloud, database and SaaS is a permanent engineering tax that grows with the surface area, not the revenue.
None of this means Cyera is unhealthy — we have no visibility into its actual gross margins, and we're not asserting they're bad. We're saying the shape of the business points to structurally higher cost-of-goods than a pure control-plane vendor (an SSO or an MFA company processes tokens, not petabytes). When a company with that cost shape raises $1.2B in nine months, "high COGS, heavy data-processing spend, or elevated burn" is the most parsimonious explanation for the appetite — and it's the thing a buyer's procurement team should probe.
Cyera has made roughly six acquisitions, including Oasis Security for about $1B. M&A at that clip is itself a use of cash — and integrating six teams, six codebases and six roadmaps is an operating cost that shows up as burn long before it shows up as synergy. A company buying growth and building coverage simultaneously will consume capital fast even if the underlying SaaS margins are fine. That's not a criticism; it's arithmetic. But it means the raise cadence can be explained without any ARR problem at all — which is exactly why the operational-efficiency question, not the growth question, is the right one to ask.
Cyera is a strong product in a real category; nothing here says don't buy it. It says buy it with your eyes open on two fronts. First, price durability. A company raising this often is either scaling into enormous demand or funding a cost structure that hasn't yet turned the corner — and you can't tell which from the outside, so negotiate as if either is true: multi-year price protection, defined overage terms, and clarity on what happens to your rate card if the funding environment tightens. Second, interrogate the cost model in your own deployment. Because the platform's cost scales with your data volume and scan frequency, ask exactly how your bill behaves as your estate grows, and get it in writing. The same economics that make this an expensive business to run are the economics that can make it an expensive product to operate.
everyone's governing the robots; almost no one has finished governing the humans
The entire market has rotated toward non-human and agent identity in the space of a year. It's the right frontier. But there's a quiet error in how it's being sold: as if human identity governance were a solved, boring, legacy problem you can skip past on the way to the exciting agentic stuff. It isn't. It's the substrate. And the incidents prove it — Hugging Face being the cleanest example.
Count the launches in this issue alone: Okta for AI Agents, agents as first-class identities, shadow-agent discovery, kill switches. Every vendor has an NHI story and an agent story. Almost none of them are leading with the unglamorous question of whether their customers' human identity programs — joiner-mover-leaver, least privilege, access certification, privileged access, offboarding — actually work. The implicit pitch is that agent governance is a new, separable layer. It is not separable. Every meaningful control you want to apply to an agent or an NHI is a control you first have to be able to apply to a person:
• Ownership. An NHI or agent is only governable if a human owns it and that ownership is maintained through the human's own lifecycle. Orphaned service accounts are just NHIs whose human owner left and whose access review never caught it.
• Least privilege. You cannot scope an agent's entitlements correctly if you never mastered scoping a person's. The same entitlement model, the same role design, the same review cadence carry over — or fail to.
• Lifecycle. Agents and tokens need creation, rotation, and revocation on a schedule. That is joiner-mover-leaver applied to a non-human — the exact muscle a mature human IGA program already has, and an immature one does not.
• Certification. "Who should have this access, and do they still need it?" is the same question whether the subject is an employee, a contractor, a service account or an agent. If your human access reviews are rubber-stamped, your NHI reviews will be too.
The Hugging Face token exposures are held up as an NHI/secrets story — leaked API tokens, exposed access to models and organizations. And at the surface layer, yes, the artifact that leaked was a machine credential. But step back one level and the root cause is human identity governance, or the absence of it:
Tokens with broad, standing scope were issued and then not governed. There was no strong lifecycle behind them, no owner accountable for each one through their own tenure, no least-privilege discipline constraining what a single leaked token could reach, and no review process that would have caught over-permissioned or stale credentials before an attacker did. Those are not gaps in an "agent governance" product. They are gaps in a foundational identity security program — the human-anchored discipline of ownership, scoping, lifecycle and review that a machine credential is supposed to inherit. The token leaked because the house wasn't in order, not because the industry lacked a shiny NHI dashboard.
Put plainly: Hugging Face was exploited because it did not have a real identity security program in place, and the machine credentials sat on top of that void. A world-class agent-governance tool bolted onto that same void would have leaked the same tokens. You cannot govern the non-human layer to a standard you haven't reached on the human layer, because the non-human layer inherits its controls, its ownership, and its lifecycle from the human one.
Sequence it correctly. Before you fund an NHI-discovery or agent-governance initiative, run an honest audit of the human foundation: Do you have accurate joiner-mover-leaver? Is least privilege real or aspirational? Are access certifications meaningful or rubber-stamped? Is there a named owner for every privileged and service account, maintained as those owners come and go? If the answer to those is shaky, spend there first — not because agents don't matter, but because agent governance built on a broken human foundation is a more expensive way to fail. The vendors selling you the agentic layer would rather you not ask this, which is exactly why you should.
a solid product inside an ownership structure we think is about to change
Per Bloomberg reporting, Clearlake Capital is weighing options for the identity assets it controls, including merging One Identity with RSA's SecurID business into a new company and selling a stake to another private-equity firm. Nothing is confirmed and no transaction has been announced. But the direction of travel is exactly what the ownership history would predict.
One Identity was carved out of Dell Software and sold (with Quest) to Francisco Partners and Elliott Management in 2016, then sold again to Clearlake Capital in 2021 as part of the ~$5.4B Quest Software deal. In May 2025, Clearlake put a reported ~$350M into a debt refinancing for Quest, and a new CEO, Praerit Garg, took over the same year. That is a lot of financial engineering and leadership change for one asset in a short window.
One Identity spans four fairly distinct products: Identity Manager (enterprise IGA, deep and complex), Safeguard (PAM), Active Roles (AD/Entra management), and OneLogin (access management/SSO, acquired 2021). Each has its own buyer, its own competitors, and its own natural acquirer. That breadth is a selling point operationally — and a liability structurally, because a portfolio of separable assets is the easiest kind to break apart and sell in pieces.
Identity Manager is a legitimately capable enterprise IGA with strengths in complex, on-prem-heavy and hybrid environments where the newer cloud-native tools are thinner. Safeguard is a credible PAM. The company reports ~7,500 customers and ~125M identities under management. This is not a distressed product; it's a solid, somewhat-legacy portfolio with a loyal base — which is precisely the kind of asset PE likes to buy, optimize, and sell.
The product is a solid B; the ownership is the variable that should move your decision. One Identity sells a broad, capable identity portfolio — deep IGA in Identity Manager, credible PAM in Safeguard, useful AD tooling in Active Roles, and access management in OneLogin. In a hybrid or on-prem-heavy enterprise, that breadth is real value that the cloud-native challengers can't always match. The scores reflect that: good product, moderate implementation burden (Identity Manager is powerful and not simple), and a loyal base.
What drags the composite is stability and roadmap confidence, and that's deliberate. An asset on its third PE owner, freshly refinanced, with a new CEO and now a reported merge-and-sell exploration, is an asset in motion. The financial owner's mandate is to exit. Everything about the current setup — the debt refi, the leadership change, the separable portfolio, the reported RSA combination — reads like preparation for a transaction, not a decade of patient independent product investment. For a buyer, that's not a reason to walk away; it's a reason to price the risk in.
Our call: within roughly the next 12–24 months, One Identity gets flipped to another PE firm, merged (most plausibly with RSA's SecurID into a new identity entity, per the Bloomberg reporting), or carved up and sold in pieces. We think a flip-or-merge is more likely than a carve-up in the near term — the RSA combination is the specific structure reported, and combining two mature identity bases to sell a stake is a cleaner PE play than piecemeal disposal. But the portfolio's modularity keeps the carve-up path live: Safeguard, OneLogin, Active Roles and Identity Manager each have distinct natural buyers.
This is a prediction, not reporting. We have no non-public information and no knowledge of any actual process beyond what Bloomberg has reported. It is our stated opinion based on the ownership history, the base rates for serial-PE-owned software assets, the recent refinancing and CEO change, and the reported RSA exploration. It may not happen, or may happen differently. For a buyer the practical takeaway is the same regardless: negotiate change-of-control protections now.
a category, tool, or idea worth knowing this fortnight
When a vendor cites membership in a security "alliance" or "blueprint" as evidence of integration, treat it as marketing until proven otherwise. Ask the only question that matters: is there a jointly-supported, jointly-accountable integration with a name, a SKU and a single support path? If yes, reward it. If it's an API integration that already existed plus a logo lockup, it earns nothing on your scorecard. Keep the press release and check it against reality in twelve months.
The Cyera analysis generalizes. When a platform's cost-to-serve scales with your usage — data volume scanned, events processed, inference run — a vendor's fundraising tells you the category is capital-hungry, and your contract needs to account for it. Ask exactly how your bill behaves as your estate grows, get overage terms in writing, and secure multi-year price protection. Capital intake upstream tends to become pricing pressure downstream.
Before approving an NHI-discovery or agent-governance project, audit the human basics that the non-human layer inherits: joiner-mover-leaver accuracy, real least privilege, meaningful access certification, and a named owner for every privileged and service account. If those are shaky, spend there first. Agent governance built on a broken human foundation — the Hugging Face pattern — is just a more expensive way to leak the same credentials.
one line to sound three moves ahead in your next exec meeting
The board will hear "AI agents" and "non-human identity" from every vendor this year. The mature framing — the one that separates you from the room chasing the shiny layer — is that agent security is downstream of identity fundamentals, not a replacement for them. Every control we want over an agent is a control we first have to run reliably over a person. So our investment sequence is foundation first, frontier second: get human identity governance genuinely operating, and the agent layer inherits controls that actually work. Skip that, and we're buying a lock for the window while the front door stands open.
a spicy anonymized take from the community this fortnight