Juniper Advanced Threat Prevention in 2026: What Replaced Sky ATP and How It Fits the Cloud-First SRX

From Sky ATP to Juniper ATP Cloud

Juniper's original 'Sky ATP' service launched in 2016 as a cloud sandbox and threat feed for SRX Series firewalls. The name is now retired: the service has been rebranded and re-architected as Juniper Advanced Threat Prevention Cloud, and integrated into the wider Juniper Security portfolio alongside Juniper Secure Edge, Juniper Secure Analytics, and the SRX and MX enforcement points. In 2026, if a vendor or blog is still calling it Sky ATP, they are two rebrands out of date, and the underlying platform is materially different from what shipped a decade ago.

The core promise is the same as any modern network sandbox: files and URLs traversing the network are extracted at the enforcement point, sent to a cloud service for static and dynamic analysis, and given a verdict that can be enforced retroactively. What has changed is the depth of integration with Juniper's routing and SD-WAN footprint, the addition of machine-learning-driven detonation and encrypted traffic inspection heuristics, and the operational tie-in with Mist AI and Marvis for correlating security events against network health.

This piece walks through how ATP Cloud actually works in 2026, where it sits relative to competing services from Palo Alto, Cisco, Check Point and Fortinet, and what to look for when you are evaluating it as part of an SRX or SASE refresh.

The pipeline: extraction, detonation, enforcement

In a typical SRX deployment, ATP Cloud is enabled per security policy. When a matching flow carries a file — HTTP, HTTPS with TLS inspection, SMTP, IMAP, FTP, SMB — the SRX extracts the file, checks its hash against Juniper's known-good and known-bad lists, and, if the hash is unknown or ambiguous, uploads the file to the cloud service. The service runs a combination of static analysis, machine-learning classification, and dynamic detonation in instrumented Windows, Linux, and macOS sandboxes.

The verdict — clean, suspicious, or malicious with a confidence score — is returned to the SRX and to the Security Director / Juniper Secure Analytics dashboard. If the verdict is malicious and the file has already reached the endpoint, the platform generates a retroactive alert. Because the SRX also participates in Juniper's identity-aware policy model, that alert is tied to a user and host, not just an IP.

The other half of the service is URL and command-and-control feeds. ATP Cloud continuously updates the SRX with reputation categories for known-malicious domains and callback destinations, so much of the value never involves a sandbox at all — it is reputation-based blocking of the traffic that would have delivered or exfiltrated in the first place. This is the workhorse capability and is generally where teams get the biggest measurable reduction in incident volume.

Encrypted traffic, TLS inspection, and the honest limits

The industry-wide problem for any network sandbox in 2026 is that the majority of malware delivery now happens over TLS 1.3 with encrypted client hello, and a growing share of enterprise traffic is genuinely opaque to middleboxes. Juniper's response, like everyone else's, is a mix of selective TLS inspection where legally permissible, encrypted traffic heuristics that look at JA4 fingerprints and flow statistics, and tight integration with endpoint telemetry so that verdicts can be made on decrypted data at the host rather than in the network.

Be realistic about what this means operationally. On a network where you are not doing TLS decryption — because of regulatory constraints, privacy commitments, or the operational cost of certificate injection at scale — ATP Cloud's file-based detonation capability is materially reduced. Reputation and behavioural detection still work, but you are relying on flow-level indicators rather than payload inspection. That is not unique to Juniper; it is true of every network-based sandbox on the market. The correct architectural response is to pair network ATP with a competent endpoint detection and response tool rather than pretending the network can see everything.

For customers doing selective inspection, Juniper's guidance and default profiles have improved. The service tags high-risk categories (unknown, recently registered, uncategorised) for inspection while leaving healthcare, finance, and government traffic decrypted only where policy allows, which is much closer to what an auditor expects than a blanket 'inspect everything' approach.

Integration with Mist, Marvis, and Juniper Secure Edge

The most meaningful change in the last two years is the integration between ATP Cloud verdicts and the AI-driven operations layer inherited from the Mist acquisition. In practice, a malicious file verdict that hits an SRX behind a Mist-managed campus can be correlated automatically with the specific WLAN, switch port, and device that generated the flow. Marvis, the AI assistant, will surface the incident in natural language and offer suggested containment actions — quarantine the client, revoke the 802.1X session, or trigger an EDR isolation via API.

For customers using Juniper Secure Edge as the SASE forwarder for remote and roaming users, the same ATP Cloud service applies to traffic that never touches an on-premise SRX. The verdict pipeline is unified; the enforcement point changes. This matters because it means an organisation moving from a datacentre-centric SRX estate to a SASE model does not have to re-buy or re-tune its threat prevention layer.

Where the integration is still catching up is with third-party endpoint and SIEM ecosystems. The APIs are there, but out-of-the-box correlation is strongest inside the Juniper stack. If you are running Juniper networking with a non-Juniper EDR and a non-Juniper SIEM, budget time for integration work rather than assuming the vendor slides.

How it compares to Palo Alto WildFire, Cisco, Check Point and Fortinet

WildFire from Palo Alto Networks remains the most widely deployed cloud sandbox, and its detection volume is unmatched simply because of install base. Cisco Secure Malware Analytics (formerly Threat Grid) is deeply integrated with Cisco Secure Firewall, Secure Endpoint, and Umbrella. Check Point's ThreatCloud AI is the most aggressive marketer of AI-driven detection in this space and, on file-based tests, is broadly competitive. Fortinet's FortiSandbox has the strongest story for pure appliance-only deployments and air-gapped networks.

Juniper ATP Cloud's differentiators in 2026 are the network-fabric integration (Mist plus Marvis), the SASE story via Juniper Secure Edge, and pricing that tends to be more forgiving than the top of the Palo Alto and Cisco stacks for customers already invested in SRX and MX platforms. Raw detonation-quality benchmarks are within the noise band of the other main players; nobody has a decisive edge on file verdicts anymore. Choose on architecture fit and operational integration, not on marketing detection percentages.

One vendor-neutral point: run a proof of value with real, current samples from your incident response history and from independent sample sources. Vendor bake-offs on curated malware zoos are almost useless because every serious vendor scores in the high nineties on those. What matters is time-to-verdict on genuinely novel payloads and false-positive rate on your normal business traffic.

Buying and operating checklist

Confirm which SKU you are licensing. Juniper ATP has multiple tiers, and older documentation still refers to legacy tier names. Make sure the tier you buy includes the file types and analysis features you need — particularly for Office document macros, PDF exploits, and Linux ELF binaries if you run non-Windows endpoints.

Confirm sandbox residency. Data residency requirements — EU, UK, ANZ, US federal — are a live discussion in most regulated buyers. Juniper offers regional detonation options, but the defaults may not match your compliance posture.

Confirm your TLS inspection stance, and if you are not doing inspection, do not expect file-based detection to be the workhorse. Configure reputation and C2 blocking aggressively, layer endpoint DR behind it, and treat the sandbox verdicts as retroactive telemetry rather than primary prevention.

Confirm your incident response playbooks understand where the verdicts appear. Nothing wastes a threat prevention licence like an alert queue no analyst reads. Integrate ATP Cloud events into your SIEM and, if possible, into the case management tool where the SOC actually works.