No. A volumetric DoS floods the link and trips volume-based detectors. ICS-Sniper adds no traffic and drops under 5% of a cycle’s packets in two short, precisely chosen windows, and it leaves liveness-check probes untouched so the link still looks alive. The novelty is identifying which few encrypted packets matter without ever seeing their contents.
No. ICS-Sniper never decrypts anything. It works purely from traffic metadata (packet sizes, timing, and direction), which a VPN leaves exposed even though it hides the payload.
No. The attacker does not need to know the control logic, cryptographic secrets, or have payload access. ICS-Sniper assumes only the general properties of ICS traffic: that site-to-SCADA communication is periodic, and that each process state has its own characteristic traffic pattern.
It requires covert, persistent control of one ISP or provider-edge router on the path between the plant and its cloud-hosted SCADA. Publicly disclosed Cisco and Juniper router vulnerabilities grant exactly this foothold, and search engines like Shodan can locate the exposed devices. These are realistic capabilities for a rogue ISP insider or a nation-state actor. However, gaining that foothold is non-trivial and is not something our released artifacts provide.
No. The reference SWaT plant’s OT network is air-gapped and not Internet-connected, so the attack cannot be turned against it. All experiments were run on lab testbeds. One has software-simulated controllers, another has real industrial PLCs, and both have simulated mechanical components. No experiment touched a live plant.
Water treatment is our case study because SWaT is one of the very few openly available, realistic multi-process control logic. The technique itself depends only on periodic request-response traffic, so it applies to the broader class of batch-processing ICS, including water, wastewater, and discrete manufacturing, and generalizes across OT protocols (Modbus, ENIP, and IEC 61850 MMS, including a power-grid dataset).
Cloud-hosted SCADA with site-to-SCADA traffic carried over a VPN is a recent but rapidly growing architecture, as confirmed by multiple vendors and industrial security practitioners. Plants don’t publish their network designs for security reasons, so no exact count exists, but the SANS 2024 State of ICS/OT Cybersecurity survey documents a surge in adoption, with water and wastewater utilities across the US, Canada, Mexico, Australia, and New Zealand among the early adopters. The exposure grows as more plants move SCADA to the cloud. Therefore, we think it is important to fix the new vulnerabilities before more widespread adoption.
The exposure exists only once SCADA traffic crosses the public Internet. A fully on-premise or air-gapped SCADA is outside this threat model.
Not as presented. ICS-Sniper relies on the difference between a batch plant’s active and idle phases. Continuous processes lack those distinct phases, which is a genuine limitation and part of our future work.
On our software-PLC testbed, one attack delayed the start of purified-water distribution by about 15 minutes; another caused a loss of already-treated water for 9 minutes 11 seconds. Such stalls delay clean water supply, waste treated water, and slow down production. In a real deployment, these effects would ripple outward to downstream facilities and the communities that depend on the supply.
Those are the four standard NIST/AWS safeguards, and ICS-Sniper evades all of them. It reads only the metadata the VPN leaves exposed; the detectors are built for volume spikes, not a handful of well-timed drops. In our evaluation against an ensemble of eight state-of-the-art ICS detectors, none of the detectors flagged the attack before the plant’s performance had already degraded.
Network detectors are event-driven. They fire on packet arrival, not on a brief silence, which is what a dropped burst looks like. And ICS-Sniper introduces no spoofed ARP, no altered payloads, and no extra traffic for conventional defenses to key on. Catching it would likely require actively polling the link in step with each PLC’s rhythm, which is an open research problem rather than a setting on an existing tool.
Because the attack needs a compromised on-path device to function, hardening routers and monitoring the ISP path still matters. Beyond that, the paper points to three directions: shaping traffic to hide the periodic pattern, diversifying the network path so one compromised hop can’t blackhole the flow, and rhythm-synchronized liveness polling. None is fully solved yet.
We weighed the risks using the Menlo Report's beneficence principle and concluded the defensive benefit outweighs the residual risk. There was no vulnerability to disclose confidentially: ICS-Sniper isn't a single-vendor bug but a weakness of the architecture itself—metadata that leaks through VPN encryption, combined with already-public router CVEs. Publishing now, while cloud-hosted SCADA is still early in adoption, gives operators and vendors time to build defenses before these systems become widespread. We've also shared the findings with the security and ICS communities, including at the National Cybersecurity Consortium.
We release the attack code, the detector ensemble, implementation details of two types of testbeds, and the captured network traces, so others can reproduce our results and build detection and mitigation techniques on top of them. The safeguards are contextual. All experiments were run on lab testbeds. One has software-simulated controllers, another has real industrial PLCs, and both have simulated mechanical components. No experiment touched a live plant. Running the attack in the wild still requires covert control of an on-path router, which the artifacts do not provide.