Between Monday’s report and Thursday, two developments stand out. One is urgent and immediately actionable. The other is architecturally important and not yet fully understood. They deserve separate attention.

The architecturally important one first: a BGP hijack targeting Softaculous update traffic delivered a backdoored build of Virtualizor, a hypervisor management platform used across hosting providers, to an unknown number of installations over a 33-hour window ending August 30. This is not a server compromise or a credential theft. This is an attacker who manipulated internet routing, obtained a valid TLS certificate for the hijacked domain, and exploited the absence of cryptographic package verification in Virtualizor’s update client to silently install a rootkit on hosting hypervisors. AlbaHost confirmed 5 of its 34 hypervisor nodes were compromised. The full scope across all Virtualizor deployments worldwide is unknown.

The Trust Chain That Broke

The Virtualizor compromise succeeded because it broke three independent trust mechanisms simultaneously, and each failure enabled the next.

From August 28 at 20:57 UTC through August 30 at 06:10 UTC, threat actors used AS62390 (NexonHost) to announce a more-specific BGP prefix covering Hetzner address space. BGP’s longest-prefix-match rule meant traffic destined for Softaculous update infrastructure was routed to attacker-controlled systems instead. This is the entry point, but it only works if the victim cannot detect the diversion.

The attacker anticipated that. They obtained a Let’s Encrypt certificate for the hijacked domain before diverting traffic. Virtualizor update clients received a valid TLS handshake from what appeared to be a legitimate Softaculous endpoint. No certificate error. No user warning. The connection looked correct because, from the certificate’s perspective, it was issued to whoever controlled that address at the time of issuance, which was the attacker.

What made the attack complete was Virtualizor’s update client lacking cryptographic package verification. Even if an operator had been watching network traffic carefully, the update would have appeared as a normal, TLS-secured software download from the expected domain. The only layer that could have caught the trojanized build was a signed package check against a known-good key. That check did not exist.

The payload implanted attacker SSH keys in root accounts, installed a persistent systemd service named java-jre-update.service, created an unauthorized proxyuser account, and beaconed to two C2 endpoints: cdn.nerat.cc and connect.ne-rat.xyz. Any hypervisor node that received the trojanized update between those timestamps and has not been reimaged should be treated as fully owned. The persistence mechanism and root-level SSH access give the attacker continued access even after the C2 connection is severed.

The hosting stack implication is the part that does not appear in most coverage. A single Virtualizor master controls hundreds of VPS nodes. A compromised hypervisor is not one compromised server; it is the management plane for every virtual machine running on that hardware. The attacker’s initial foothold is at the layer above all those tenants. Softaculous has released Virtualizor 3.2.9.9 with a remediation tool and is implementing code signing for future updates, but the scope question, how many installations beyond AlbaHost received the malicious build, has not been answered publicly.

The defense lesson is not primarily about BGP monitoring, though that is worth addressing separately. It is that software update integrity verification is a binary control. You either cryptographically verify packages against a published key, or you accept the risk that a sufficiently positioned attacker can deliver arbitrary code through your update channel. TLS alone does not protect update integrity; it only protects the transport layer, which the attacker controlled.

SonicWall SMA1000: The Third Time This Year

The immediately actionable development: SonicWall disclosed two actively exploited zero-days in SMA1000 remote access appliances on September 2. CVE-2026-83548 (CVSS 10.0) is a pre-authentication SSRF in the WorkPlace interface. CVE-2026-83549 (CVSS 7.8) is an OS command injection in the Appliance Management Console. Attackers are chaining the pair for unauthenticated remote code execution, achieving full system control without any credentials.

Affected models are 6210, 7210, and 8200v. Hotfixes are available in versions 12.4.3-03526 and 12.5.0-02952. SonicWall’s guidance includes reimaging compromised appliances and rotating all credentials and TOTP tokens.

The pattern matters as much as the specific CVEs. This is the third time SMA1000 infrastructure has been actively exploited in 2026. Remote access appliances have been a preferred initial-access target across the industry for two years, but when a single product line sees three separate exploitation events in eight months, it raises a question about whether the attacker pool targeting this class of device has expanded or whether there is a specific reason SMA1000 keeps appearing. At this point, organizations running SMA1000 should apply the hotfix, treat any appliance in the affected version range as potentially compromised regardless of visible indicators, and assess whether the architecture warrants a longer-term replacement decision.

Escalation from Monday: The ICS Exploitation Barrier Is Measurable Now

Monday’s brief covered the August 19 joint CISA/NSA/FBI advisory warning that threat actors are using AI-generated Python scripts to target Siemens S7 PLCs. The advisory was important but remained at the level of observed activity without a quantified capability measurement.

That measurement arrived this week. Forescout Vedere Labs published research demonstrating an AI-assisted workflow that ported a working RCE exploit from a WAGO 750-852 PLC to a WAGO 750-831, a different model without source code or debugger access, in 8 hours and 32 minutes at a cost of $535.74 in API calls. The starting point was CVE-2021-31886, a known flaw, but the target model had never been publicly exploited. The work required genuine technical judgment from the AI system, not just template substitution.

Paired with eight CISA ICS advisories released September 1 covering Mitsubishi Electric FA Engineering Software and seven Rockwell Automation product lines, the picture is consistent: the ICS advisory volume is elevated, the offensive tooling barrier is compressing, and the combination is not a coincidence. The Forescout research provides a concrete cost floor for OT exploit development that did not exist publicly before this week. Security teams arguing for continued deferral of OT patch cycles based on “specialized attacker requirements” are now arguing against a published $536 counterexample.

What to Watch

The Virtualizor scope question will define how significant this incident becomes. If the BGP hijack window reached only a handful of installations, it is a serious but bounded event. If Virtualizor’s update distribution is centralized through the affected Hetzner/Softaculous infrastructure, the compromised node count could be substantially larger than what AlbaHost disclosed. Watch for additional hosting providers confirming compromise in the next 72 hours. The java-jre-update.service indicator is specific enough to be hunted across all Virtualizor deployments; anyone running this platform should check for it now.

On SonicWall, attribution will tell us whether this is opportunistic scanning by a broad attacker pool or targeted activity consistent with prior SMA1000 campaigns. Three exploitation events on the same product line in one year warrants tracking that question actively.

Security Unlocked publishes threat intelligence and strategic analysis twice weekly. This mid-week brief covers developments from August 31, 2026 through September 3, 2026.