CASB Vendors in 2026: Who Actually Sees Data Leaving?

Short answer: Every CASB vendor picks one of two places to stand, and neither one is where your data actually leaves. API-only CASBs connect to a SaaS tenant and scan files after they land. Proxy CASBs inspect traffic, but only after it has detoured to the vendor's nearest data center. dope.security runs CASB and DLP on the device itself, so a file heading to an unsanctioned AI tool gets inspected at the moment it leaves the laptop, not after it arrives somewhere you happen to have an API key.

That is the whole argument of this post, and it is a falsifiable one. If your CASB cannot inspect a file on its way to a personal ChatGPT account on an unmanaged network, it is not covering the egress path that matters most in 2026. Most of the category cannot.

If you are still working out where the category boundaries sit, our breakdown of how CASB and DLP actually differ is the right place to start. This post assumes you already know you need one and are trying to work out which CASB vendors are worth a shortlist.

What a CASB is supposed to do, and where each vendor puts it

A Cloud Access Security Broker sits between your people and your SaaS applications. It is supposed to answer three questions: which cloud apps are in use, what data is moving into and out of them, and whether the account doing the moving belongs to the company or the person. Every CASB vendor claims all three. The differences show up entirely in architecture.

There are two ways to build one. API mode connects to sanctioned SaaS tenants through their own APIs and inspects data at rest. It is accurate, it is easy to deploy, and it is always looking backward, because a file has to already be in the tenant before the API can see it. Inline mode puts the CASB in the traffic path so it can act before the upload completes. That requires a proxy, and a proxy has to live somewhere.

For most vendors, somewhere means a point of presence. Your laptop sends the request to a data center, the data center inspects it, and the data center sends it on. On the way back, the same detour happens again. That is the design decision that shapes everything else about the product, including what it costs and what it misses.

The CASB vendor landscape, vendor by vendor

These are the CASB vendors that show up most often on a mid-market shortlist. Each one scores well on something. The point below is not that they are bad products, it is that the inspection point decides what they can and cannot do.

Netskope

Netskope has the richest CASB feature set on paper and it is the vendor most likely to already do what you are asking for. Inline inspection runs through NewEdge, Netskope's own proxy network. Netskope's published SLA commits to under 10 ms of added latency for non-decrypted traffic and 50 ms once traffic is decrypted, which is a fivefold penalty on exactly the traffic you care about inspecting. The API-based CASB is licensed as a separate SKU from the inline product, and full inline DLP plus threat plus AI controls sit in the higher Max Advantage tier. Customers and reviewers consistently report that it is hard to deploy and administer, and that cert-pinned applications end up on bypass lists.

Zscaler

Zscaler's CASB is part of the Zero Trust Exchange, which means all traffic forwards to a ZEN or Service Edge node before it goes anywhere. Gartner has cited a 10 to 20 percent throughput drop as inspection modules stack, and latency compounds as you turn more of them on. Zscaler's own documentation puts prompt-level data protection behind the Data Protection add-on, with AI Guard and AI Scanning licensed separately again. The architecture cannot inspect cert-pinned Microsoft 365, WebEx or Dropbox traffic, and developer tooling such as Docker, Python and Git commonly breaks under inspection.

Skyhigh Security

Skyhigh is the original CASB, carved out of McAfee, and it still carries the deepest heritage in the category. It is also the vendor Gartner was least kind to. The 2024 SSE Magic Quadrant placed Skyhigh as a Visionary rather than a Leader and noted that the McAfee and Trellix split disrupted its sales motion, with no significant advancement that cycle. Its deeper controls for AI tools require enterprise or API access on sanctioned tools, which means personal AI accounts fall back to coarse URL and upload control.

Forcepoint ONE

Forcepoint's CASB came from the Bitglass acquisition and runs a reverse-proxy model that Forrester described as a double integration problem alongside Skyfence. It backhauls through roughly 300 AWS points of presence, and customers report policy changes taking 20 to 30 minutes to take effect. Forcepoint's own knowledge base confirms that its offices in China are blocked, which matters if you have people there.

Cloudflare One

Cloudflare has the fastest edge in the group and the youngest SSE stack. Its inline CASB covers roughly 25 application categories, and unlimited CASB, full DLP, remote browser isolation and long log retention are all gated to the Contract plan. AI Prompt Protection shipped in beta in August 2025 covering about four applications, with header-based tenant control limited to Google and Microsoft. The November 18 2025 outage, caused by a single oversized configuration file, produced global 5xx errors for roughly five hours and took third-party services down with it, which is what uniform anycast buys you when there is no regional isolation.

dope.security

dope.security runs the CASB on the endpoint. CASB Neural handles data at rest, scanning Google Workspace and Microsoft 365 for publicly or externally shared files containing PII, PCI, PHI and intellectual property, with one-click remediation and no policy tuning before it starts. Dopamine DLP handles data in motion on the device, intercepting file uploads and AI prompts and classifying them through zero-retention APIs, under US Patent 12,464,023. Cloud Application Control adds tenant-level enforcement, so the corporate ChatGPT account works and the personal one does not, on the same domain. All of it runs from one console in under 100 MB of RAM.

The comparison that actually decides this

Strip away the feature matrices and the difference comes down to a short list of behaviors. Here is the same question asked of both architectures.

• Inspection point: proxy CASBs inspect in a data center your traffic has to reach first; dope.security inspects on the device before the request leaves.
• Data at rest and data in motion: most vendors sell API CASB and inline CASB as separate SKUs; dope.security ships CASB Neural and Dopamine DLP under one console and one agent.
• Personal versus corporate accounts: tenant control at most vendors needs the proxy plus a data-protection add-on plus a higher tier; dope.security enforces tenant restrictions on the device as a native capability.
• Cert-pinned applications: cloud proxies break pinned apps and force blind bypass lists; dope.security inspects on the endpoint where the certificate store lives, and surfaces SSL errors so admins create bypasses in a few clicks rather than guessing.
• Restricted geographies: several vendors sell China connectivity as a paid uplift or run through a local partner; dope.security works there without a separate SKU because there is no point of presence to reach.
• Failure mode: when a cloud control plane goes down, customers lose dashboards and logs mid-incident; when inspection is local, the agent keeps enforcing cached policy.

The AI problem the CASB category was not built for

CASB was invented to solve shadow SaaS. An employee signs up for a file-sharing tool, the CASB spots the domain, IT either sanctions it or blocks it. That model assumes the risky app is a website with a tenant you can connect to.

AI broke both assumptions. The risky surface is now a prompt box, and the data leaves as text rather than as a file. Worse, the same domain serves both the sanctioned and the unsanctioned version. Allowing the corporate ChatGPT workspace while blocking personal accounts needs an HTTP header inspected inside decrypted TLS. DNS cannot see it. A browser extension only sees the browser. An API CASB never sees it at all, because the personal account is not your tenant.

This is why shadow AI is a different discovery problem from shadow IT, and why AI clients that are not browsers (desktop apps, IDEs, command line tools) fall outside most of the category entirely. If your CASB shortlist cannot answer the corporate-versus-personal question on the same domain, it is solving the 2019 version of the problem.

What the detour costs, in real numbers

Every inline CASB that is not on the device adds a round trip. Measured cloud-proxy latency runs roughly 40 to 80 ms when a user sits near a point of presence and 150 to 400 ms when they do not. That is per request, and a business application chains a dozen or more requests before a page is usable. Netskope's own SLA concedes the decrypted number is five times the non-decrypted one.

Want to see what the detour costs on your own connection? The Fly-Direct Speed Test measures your real round-trip latency in the browser and shows app-by-app load times against a legacy cloud proxy. Read how Fly Direct inspection works on the device, or book a 20-minute demo and we will run it with you.

The takeaway: inspection has to happen somewhere, but it does not have to happen somewhere far away.

We have written the arithmetic out in full in our breakdown of what cloud proxy latency actually adds up to, and the architectural version in the backhaul bottleneck.

What to ask a CASB vendor before you sign

Five questions separate the shortlist quickly, and all five have documented answers rather than opinions.

• Which SKU do inline CASB, API CASB and DLP each belong to, and what is the price at renewal rather than at signing?
• Can you allow a corporate SaaS tenant and block a personal one on the same domain today, and what does that require?
• What happens to policy enforcement when your control plane is unavailable?
• Which cert-pinned applications end up on the bypass list, and what is the process for adding one?
• What is the added latency for decrypted traffic, in milliseconds, in your own SLA?

Where this leaves the shortlist

If your people work in one building and your SaaS estate is small and fully sanctioned, an API-only CASB is a reasonable and cheap answer. If your people are distributed, if AI tools are already in the building whether or not you approved them, and if you care about stopping data rather than reporting on it afterward, the inspection point has to move to the device.

That is the bet dope.security made. Greylock Partners went from first proposal to signed contract in 27 days, and a Fortune 100 customer scaled from 900 devices to more than 18,000 in weeks, roughly 3,000 per week, deployed silently through Intune. Neither of those numbers is available to you when deployment means standing up traffic steering to somebody else's data center.

Ready to see it? Start a free CASB Neural scan and find out which files in your tenant are already publicly shared, or book a 20-minute demo.

Frequently Asked Questions

What is the difference between a CASB and a DLP?

A CASB governs access to cloud applications and gives you visibility into which apps are used and by whom. DLP inspects the content moving through them and decides whether it should move. The two overlap heavily in 2026, which is why dope.security ships them together: CASB Neural for data at rest in Google Workspace and Microsoft 365, and Dopamine DLP for data in motion on the device.

Do I need a CASB if I already have a secure web gateway?

Only if the two are genuinely one product. A SWG sees web traffic and a CASB understands cloud application context, and when they come from different vendors or different SKUs you end up reconciling two policy models. dope.security builds the SWG, the CASB and the DLP on a single agent and a single console, so an application decision and a data decision are made in the same place.

Is an API-based CASB enough on its own?

Not for anything that has to be stopped rather than reported. API CASBs inspect data after it has landed in a tenant you control, which makes them blind to uploads into personal accounts, unsanctioned tools and AI prompt boxes. They are useful for governing sanctioned SaaS and they should be paired with inline inspection that runs before the upload completes.

Which CASB vendors can block personal ChatGPT while allowing the corporate account?

Doing this requires reading an HTTP header inside decrypted TLS on the same domain, so DNS-layer and browser-only products cannot. Among the proxy vendors it is generally possible but usually requires the proxy plus a data-protection add-on plus a higher tier. dope.security does it on the device through Cloud Application Control, without an add-on.

Why is CASB pricing so hard to compare?

Because most vendors split the category across modules. Inline CASB, API CASB, DLP, threat protection and AI controls are frequently separate SKUs or separate tiers, so the quote you compare at signing is rarely the bill you get at renewal. Ask which specific SKU each capability lives in before you compare numbers.

Do CASB products work in China and other restricted regions?

Several proxy vendors sell China connectivity as a premium uplift or operate through a local partner with an ICP filing, and Forcepoint's own knowledge base confirms its China offices are blocked. dope.security inspects on the device, so there is no point of presence to reach and no separate China SKU to buy.