# Most published SASE SLAs promise availability and pay credits, and few promise latency or packet loss

Guide, 12 Nov 2026, read 12/13 Nov 2026

**The short answer.** Published SASE SLAs differ more in their exclusions than in their headline number. Barracuda commits 99.5% per month, Forcepoint 99.99%, Zscaler 99.999% and iboss 99.99999% on commercially reasonable efforts. Credits are the only remedy in the Zscaler, Forcepoint, iboss and Versa texts read on 12 Nov 2026.

## Start with what the number allows

Convert every availability figure to minutes before you compare. In a 30-day month there are 43,200 minutes. At 99.5% the service can be down about 3.6 hours. At 99.99% it can be down 4.3 minutes. At 99.999% it can be down about 26 seconds. At 99.99999% it can be down about a quarter of a second, which no operator can measure in practice.

That last figure is the one iboss's SLA text uses, as a commitment to commercially reasonable efforts. The same company's web pages state 99.999%. When two documents from one vendor disagree, the contract document governs, so I would ask which one goes in the order form.

## Terms side by side

These are the SLA documents I could read on 12 Nov 2026. Each row is the vendor's published text, not a measured result.

Notice that no row has all three of availability, latency and packet loss with credits attached. Aryaka comes closest, and it caps all credits at 25% of monthly fees.

| Vendor | Availability and quality terms | Credits and limits |
| --- | --- | --- |
| Zscaler | 99.999% per month for ZIA, ZPA and ZDX; ZIA latency 100 ms or less at the 95th percentile of qualified traffic | Up to 30 days of fees; sole remedy; credits are not refunds |
| Forcepoint | 99.99% for proxied traffic; latency credit if average processing latency exceeds 100 ms for more than 5% of the month | 1 day per 2-hour period below target; outages of one hour or less excluded; claim in writing within 5 days |
| iboss | 99.99999% target; 100 ms or less at the 95th percentile, measured to a gateway in the same country | Credits of 5, 10, 20, 30 or 50% of one-twelfth of the annual fee; request within 30 days; cap of 30 days of fees a year; none on terms under one year |
| Versa | 99.999% monthly for hosted SSE; processing at or under 10 ms (no TLS) and 50 ms (AV/IPS with TLS) | Credits of 2 to 24 days; ticket within 2 days of month end; applies to the SaaS data plane only |
| Aryaka | PrivateCore 99.999% and packet loss under 0.1%; 99.99% end to end with HA and dual ISP at both sites | Packet loss miss pays 10% of Link Service Fees; total cap 25% of monthly fees; future invoices only |
| Barracuda | 99.5% per calendar month per region; no latency or loss figure | Outages under 5 continuous minutes and hosting provider failures excluded |

## Latency is measured where it flatters the vendor

Zscaler and iboss both publish 100 ms at the 95th percentile. That is generous: it allows one request in twenty to exceed it and still be in compliance. iboss measures it between your network interface and a gateway in the same country, and states it is not a measure of paths between PoPs. A customer pointed at a gateway in another country gets no credit.

Versa's 10 ms and 50 ms are processing times inside the service, not round-trip times to a user. Netskope states a traffic-processing latency SLA of under 10 ms and is the only numeric SLA term found on the fetched Netskope pages. Palo Alto Networks states 10 ms security processing and 35 ms SaaS performance on its product page, but its SLA document returned HTTP 404 so I could not read the measurement terms.

## Stated but not published as terms

Several vendors state a number with no SLA text to read. Cato states a 99.999% backbone uptime SLA; the credits and measurement were not found on 12 Nov 2026. Fortinet's data sheet states 99.999% with a latency guarantee and then says nothing in it is a binding commitment. Cloudflare's Access page states 100% uptime for paid plans, but its Zero Trust SLA pages returned HTTP 404, while the Enterprise Support SLA caps credits at six months of fees for serving customer content. Open Systems states 99.999% for its backbone and 99.99% end to end with its last-mile service, with credit terms none found.

For Check Point, HPE Networking and Arista VeloCloud I found no SLA on the pages read. Cisco's Offer Description states a 99.999% availability target on commercially reasonable efforts, and the SLA text with credits is available only on request. Treat a stated number without a document as marketing until it appears in your order form.

## Exclusions that change the answer

Read the exclusions before the headline. Barracuda excludes platform and hosting provider failure, which for an Azure-hosted service is a large category. iboss excludes scheduled maintenance, force majeure and downtime from iboss-supplied hardware. Palo Alto Networks excludes lab tenants. Aryaka may update its SLA without notice, and for circuits it states that repair times follow the third-party provider.

Zscaler has a separate China Premium SLA: 99.999% availability, a mean time to repair of 5 hours and packet loss at or below 1% between mainland China and Hong Kong. It is the only published term I found that puts a repair time and a loss figure on the China leg.

## What I would put in the contract

Ask for four changes: a measured latency term between your two busiest PoPs, credits that are automatic instead of claim-based, a termination right after three missed months, and a definition of availability that includes the management plane. Forcepoint, for instance, publishes 99.9% for its Data Security portal against 99.99% for traffic, so the console you change policy in can be down longer than the traffic path.

## Credits are small money

Compare the credit to the loss. As an illustration, take an annual fee of $400,000. Zscaler's cap of 30 days of fees is about $33,000. iboss pays at most 50% of one-twelfth of the annual fee, about 4% of the annual fee, which is $16,700 on the same figure. A day-long outage across 100 branches costs most companies more than either.

So the SLA is not insurance. Its value is that it gives you a documented basis to leave and a reason for the vendor to staff the fix. Ask who measures availability, from where, at what interval, and whether the figure covers only qualified traffic, a phrase that Zscaler's text uses.

## Sources

- Zscaler, SLA and support terms: https://www.zscaler.com/legal/sla-support (read 12 Nov 2026)
- Zscaler, China Premium Access product sheet: https://www.zscaler.com/china-premium-access-productsheet (read 12 Nov 2026)
- Forcepoint, security service level agreement: https://www.forcepoint.com/legal/security-service-level-agreement (read 12 Nov 2026)
- iboss, service level agreement: https://www.iboss.com/service-level-agreement (read 12 Nov 2026)
- iboss, cloud data centers: https://www.iboss.com/cloud-data-centers (read 12 Nov 2026)
- Versa, hosted SSE SLA: https://versa-networks.com/documents/Versa-Hosted-SASE-SLA.pdf (read 12 Nov 2026)
- Aryaka, service level agreement: https://www.aryaka.com/aryaka-service-level-agreement/ (read 12 Nov 2026)
- Aryaka, services terms: https://www.aryaka.com/services-terms/ (read 12 Nov 2026)
- Barracuda, SecureEdge product and service description: https://assets.barracuda.com/assets/docs/dms/Barracuda-SecureEdge-Product-and-Service-Description.pdf (read 12 Nov 2026)
- Netskope, NewEdge: https://www.netskope.com/newedge (read 12 Nov 2026)
- Palo Alto Networks, Prisma Access: https://www.paloaltonetworks.com/sase/access (read 12 Nov 2026)
- Fortinet, FortiSASE data sheet: https://www.fortinet.com/content/dam/fortinet/assets/data-sheets/fortisase.pdf (read 12 Nov 2026)
- Cisco, Secure Access Offer Description: https://www.cisco.com/c/dam/en_us/about/doing_business/legal/OfferDescriptions/Cisco-Secure-Access-Product-Description.pdf (read 12 Nov 2026)
- Cloudflare, Enterprise Support SLA: https://www.cloudflare.com/enterprise-support-sla/ (read 12 Nov 2026)

Last reviewed 12 Nov 2026. Canonical: https://bestsasesolutions.com/guides/reading-a-sase-sla/
