Daily Cybersecurity News – September 2, 2026

What happened

SonicWall released hotfixes for two flaws in its SMA 1000 series VPN appliances after investigating a case of active exploitation that may chain the bugs for remote code execution.

CVE-2026-83548 is a pre-authentication SSRF (CVSS 10.0) in the Appliance Work Place interface that lets a remote unauthenticated attacker gain unauthorized access to sensitive functionality. CVE-2026-83549 is a post-authentication OS command injection (CVSS 7.8) in the Appliance Management Console that can lead to RCE under specific conditions when the attacker already has admin rights.

The flaws were discovered internally by SonicWall researchers. The vendor has not shared details on the exploitation activity or the actors involved. This follows earlier actively exploited SMA 1000 issues (CVE-2026-15409 and CVE-2026-15410) used to deploy malware.

Who is affected

SMA 1000 models 6210, 7210, and 8200v running 12.4.3-03453 (platform-hotfix) and older, or 12.5.0-02835 (platform-hotfix) and older.

Organizations using these appliances for remote access and VPN are exposed until upgraded. Scale is limited to deployed SMA 1000 series units still on vulnerable builds.

Why it matters

VPN and secure access appliances sit at the network edge. Successful chaining of pre-auth SSRF into authenticated command injection can give attackers full control of the device, enabling lateral movement, credential theft, and persistent access into corporate networks.

Prior SMA 1000 exploitation already showed real-world ransomware and malware deployment risk. Edge devices remain high-value targets because a single compromise can expose large internal environments.

How it could have been prevented

Upgrade immediately to 12.4.3-03526 (platform-hotfix) or 12.5.0-02952 (platform-hotfix).

Review systems for indicators of compromise. If IoCs are found, re-image or re-deploy the appliances, change all user and administrator passwords, and reset TOTP. Restrict management interfaces and monitor for anomalous traffic.

What happened

An authentication bypass in JFrog Artifactory (CVE-2026-82329, CVSS 9.8) was exploited in the wild just days after JFrog released a patch on August 28, 2026, according to watchTowr research reported around September 1.

Under default configuration the weakness lets an unauthenticated attacker with network access obtain administrative privileges. Attackers have been observed creating admin tokens on vulnerable instances.

The flaw sits in a core artifact repository used for builds, packages, and container images.

Who is affected

Self-hosted JFrog Artifactory instances running vulnerable versions prior to the August 28 patch, especially those with default configurations and network exposure.

Organizations that rely on Artifactory as a single point of trust in CI/CD and software supply chains face elevated risk. Cloud or tightly locked-down deployments may be less exposed.

Why it matters

Artifactory holds finished code and is a choke point in automated deployment pipelines. Admin access lets attackers poison builds, inject backdoors, overwrite artifacts, or exfiltrate proprietary binaries, potentially compromising every downstream system and customer that pulls from the repository.

Compromise of one repository can cascade to hundreds of victims. Supply-chain attacks of this type amplify blast radius far beyond a typical web application breach.

How it could have been prevented

Apply the JFrog patch for CVE-2026-82329 immediately on all self-hosted instances.

Revoke and reissue every administrator token on instances that ran vulnerable builds. Disable anonymous access, enforce strong admin authentication, prohibit overwriting published artifact versions, keep repositories off the public internet behind network controls, require cryptographic signing and provenance attestation, and pin production manifests to verified digests with admission controller checks.

What happened

Forescout Research - Vedere Labs used Anthropic's Claude to port a working pre-authentication remote code execution exploit from a WAGO 750-852 PLC to a WAGO 750-831 running firmware V01.04.16, executing attacker-supplied ARM shellcode on live hardware.

The underlying flaw is CVE-2021-31886, a stack-based buffer overflow in the Nucleus FTP server's USER command handling (CVSS 9.8), reachable before authentication over TCP port 21. Researchers supplied the original exploit, firmware binary, and physical target; Claude (moving from Sonnet 4.6 to Opus 4.6) adapted the sequence under interactive researcher guidance, including changing USER/QUIT to USER/CWD and omitting CRLF so the buffer survived.

Final RCE development cost about $535 in API usage over roughly 8.5 hours. A later attempt to build a C2 implant bricked the PLC by writing to flash-mapped memory. CERT@VDE notes no updates are available for the affected controllers.

Who is affected

WAGO programmable logic controllers using the vulnerable Nucleus FTP server implementation, specifically models such as 750-852 and 750-831 (and related Siemens APOGEE/other devices listed under the CVE) where FTP on port 21 remains enabled.

Industrial and OT environments that have not disabled or blocked FTP face exposure. No widespread active exploitation of this specific porting work is described.

Why it matters

The work shows large language models can accelerate exploit porting across hardware variants when steered by skilled researchers, lowering the time and expertise barrier for adapting known critical ICS flaws.

Successful RCE on a PLC can enable deep lateral movement in operational technology networks. Even without mass exploitation today, the demonstration highlights residual risk from unpatchable legacy FTP services and the dual-use potential of AI coding assistants in offensive research.

What happened

Between August 28 and August 30, 2026, a threat actor BGP-hijacked a block of Softaculous IP addresses used for software updates and other services, diverting traffic to attacker-controlled infrastructure.

AS62390 (NexonHost) announced a more-specific prefix covering part of Hetzner's space while retaining AS24940 on the path, so the rogue route took precedence. The actor obtained a valid Let's Encrypt TLS certificate for Softaculous domains via automated domain-validation that followed the hijacked path.

A malicious Virtualizor update package was delivered to a small number of installations that checked for updates during the window. Softaculous restored legitimate routing and confirmed the impact was limited to a handful of servers rather than the general user base. Investigation into other products continues.

Who is affected

Virtualizor (Softaculous VPS management panel) operators whose installations checked for updates while traffic was diverted, plus any other Softaculous services on the hijacked addresses (updates, client area, billing).

Softaculous states only a small number of Virtualizor servers received the malicious package. Broader user base was not generally affected, but operators are urged to check for compromise.

Why it matters

BGP hijacking combined with a valid TLS certificate defeats ordinary HTTPS trust and can silently deliver trojanized updates. Virtualizor manages virtual servers, so a malicious update can yield full host control, credential theft, or further supply-chain spread.

Even short-lived routing attacks against update infrastructure create high-impact compromise paths that logs on the legitimate origin never see.

How it could have been prevented

Verify Virtualizor installations for unexpected packages, persistence, or anomalous behavior and re-install from known-good sources if compromise is suspected.

Monitor BGP for unexpected more-specific announcements of your prefixes, use RPKI/ROA where possible, pin update endpoints or verify package signatures and hashes out-of-band, and restrict automatic update checks to trusted windows or authenticated channels.

What happened

Researchers at Socket identified 13 malicious Composer theme packages on Packagist that inject JavaScript into Vietnamese movie and comic streaming sites. The injected code runs mobile ad-fraud and gambling redirects and, on iPhones, a WebKit-to-kernel exploit chain that installs spyware to steal cryptocurrency wallet seeds and other data.

The packages span namespaces including vsmov, vsphim, haiau009, chilltvcms, and ophimcms. The iOS chain weaponizes CVE-2025-31277 and CVE-2025-43529 (both actively exploited WebKit issues) then escapes the sandbox via GPU process and a kernel path through AppleM2ScalerCSCDriver. Apple patched the kernel issue in iOS/macOS 26.1 (possible related CVEs include CVE-2025-43398, CVE-2025-43510, CVE-2025-43520).

The campaign continues earlier 2026 activity. A redeployed chain around August 12, 2026, mainly targeted iOS 18.4 through 18.6.x. Successful payloads collect keychain, Wi-Fi passwords, SMS, contacts, photos, cookies, call and location history, encrypt with AES, and exfiltrate to rotating C2 domains.

Who is affected

Operators of sites that installed the malicious Packagist/Composer themes (primarily Vietnamese streaming sites) and their visitors, especially users of unpatched iPhones on iOS versions vulnerable to the WebKit and kernel bugs (notably 18.4-18.6.x and earlier unpatched builds).

Crypto wallet users on those devices face direct theft risk. Site owners who pulled the themes from Packagist are the initial distribution vector.

Why it matters

Supply-chain compromise of popular CMS themes turns legitimate websites into drive-by exploit hosts. Chaining browser bugs to kernel access yields full device control and bulk theft of credentials, messages, and wallet seeds without user interaction beyond visiting the site.

Actively exploited WebKit CVEs plus a kernel escape make this a high-impact mobile spyware delivery path, particularly against users slow to update iOS.

How it could have been prevented

Site owners: remove the listed malicious packages, audit Composer dependencies, and scan for injected JavaScript. Rebuild from clean sources.

iPhone users: update immediately to the latest iOS (patches for the WebKit issues and kernel escape are in 18.6+, 18.7.3, 26.x lines as applicable). Avoid untrusted streaming sites, use strong device passcodes, and monitor crypto wallets for unauthorized activity. Packagist and similar registries should continue malware scanning of packages.

What happened

METR, a non-profit that evaluates frontier AI models on long-horizon agentic tasks, disclosed two security incidents. In March 2026 attackers stole an API key for inference on public models and consumed credits worth about $600,000 (the usage was free to METR via the provider, so no actual bill was incurred).

A researcher ran agents on a personal public-facing EC2 instance behind Google auth. A fail-open vulnerability in the vibe-coded app silently disabled authentication, exposing the orchestration dashboard. Attackers likely found it via certificate transparency and high-signal LLM/agent keywords, prompted an agent to reveal the key, added an SSH key for persistence, and burned tokens for three weeks. High normal evaluation volume and lack of spend caps delayed detection.

In May 2026 a separate financially motivated campaign systematically probed public infrastructure, including an unsuccessful attempt via an inadvertently exposed endpoint. No sensitive evaluation data is believed to have been accessed. Incidents were shared with partner AI companies before public disclosure.

Who is affected

METR's public-model inference account and the exposed researcher infrastructure. Partner AI companies received findings. No evidence of broader compromise of internal evaluation data or attribution to a named group.

Organizations running similar agent dashboards or API keys on lightly protected personal or experimental cloud instances share the same exposure pattern.

Why it matters

API keys for large model providers are high-value targets. A single exposed key can generate massive compute bills or enable abuse of models for further attacks. Fail-open auth bugs in quickly built "vibe-coded" tools turn temporary experiments into public attack surface.

The incident underscores that research and evaluation environments need the same credential hygiene, spend controls, and monitoring as production systems, especially when agents and public endpoints are involved.

How it could have been prevented

Never place organization credentials or sensitive data on non-organization infrastructure. Enforce authentication that fails closed, add spend caps and alerts on API keys, monitor for anomalous token usage, lock down experimental instances, and review certificate-transparency and newly registered domains for accidental exposure. Rotate keys after any suspected incident.

What happened

Threat actors breached the Philippines nuclear research agency by exploiting known, unpatched vulnerabilities in ownCloud file-sharing software to gain initial access.

Stolen data reportedly included reactor-related databases, personnel records, and credential stores. The attack relied on commodity flaws rather than novel zero-days, highlighting long-lived unpatched exposure on internet-facing collaboration systems.

Who is affected

The Philippine nuclear agency (and specifically its ownCloud deployment) suffered the confirmed breach. Any other organizations still running unpatched, internet-reachable ownCloud instances with the same commodity vulnerabilities share similar risk.

Impact centered on sensitive scientific, personnel, and authentication data held by the agency.

Why it matters

Nuclear and critical-research organizations hold highly sensitive technical and personnel information. Exploitation of years-old or well-known ownCloud bugs shows that basic patch hygiene failures can yield strategic data theft.

Credential stores in particular enable further lateral movement or reuse against other targets. Commodity vulnerability exploitation remains effective against under-maintained infrastructure.

How it could have been prevented

Patch ownCloud to the latest supported release immediately and remove or isolate any end-of-life instances from the internet.

Enforce multi-factor authentication, network segmentation for file-sharing platforms, regular vulnerability scanning, least-privilege access to reactor and personnel data, and monitoring for unusual download or authentication patterns. Rotate credentials if compromise is suspected and audit for persistence.