Today’s digest is about trust placed in the wrong layer — an AI tool that trusts any network caller, a VPN vendor’s own installer, and a hosting platform’s database-rename routine. Each one turns a convenience feature into a takeover path once that trust breaks. Here’s what changed since yesterday’s digest.
Vulnerability watch: CISA adds a critical, unauthenticated Langflow RCE to its KEV catalog as a Metasploit module emerges
CISA added CVE-2026-9198 (CVSS 9.8) to its Known Exploited Vulnerabilities catalog on August 5, citing evidence of active exploitation. The flaw affects Langflow OSS versions 1.0.0 through 1.10.0 and requires no authentication at all: an attacker chains the /api/v1/auto_login endpoint — which mints a SUPERUSER token for any caller on the network — with /api/v1/validate/code, which runs submitted code through exec(). Together the pair hand an unauthenticated attacker full remote code execution on a default Langflow deployment. The bug was quietly fixed in version 1.10.1 back in July, but public proof-of-concept code and a Metasploit module have both landed since, which is almost certainly why CISA moved it into KEV now. The Hacker News · BleepingComputer · Mallory
If your team spun up Langflow to prototype an AI agent or RAG pipeline, treat this the way you’d treat any internet-facing admin console: check you’re on 1.10.1 or later today, and if you can’t confirm the version immediately, pull the instance off the public internet while you check. A design where “logged in” and “SUPERUSER” are one unauthenticated request apart is exactly the kind of shortcut that’s common in AI tooling built for fast iteration rather than production exposure — audit any other internal AI tools for the same auto-login pattern before assuming this is a one-off.
Supply chain watch: a trojanized QuickFox VPN installer has quietly delivered a Mustang Panda backdoor since 2025
Researchers disclosed a long-running supply chain compromise of QuickFox, a VPN and network-acceleration tool popular with Chinese international students and expats living abroad. The trojanized Windows installer has been distributing malicious payloads since at least August 2025 — a full year of exposure before public disclosure. A modified Electron renderer file pulls in a JavaScript-based loader that fingerprints the victim’s machine before deciding whether to proceed, and on matching targets it downloads and installs FDMTP, a backdoor previously linked to the Chinese state-sponsored group tracked as Mustang Panda. QuickFox has since stripped the malicious components from its installer in version 3.59.6. Given the app’s user base, researchers suspect the campaign specifically targeted Chinese citizens residing outside China rather than a broad, indiscriminate audience. The Hacker News · Infosecurity Magazine · BleepingComputer
The victim-fingerprinting step is the detail worth remembering: a compromised installer that only detonates against specific targets will sail through casual sandboxing and most automated triage, because it looks clean to everyone else who runs it. If anyone in your organization travels frequently or supports remote staff in the region, treat consumer VPN and acceleration tools as unmanaged software — block them at the endpoint policy level rather than relying on users to self-report what they’ve installed, and update anyone still on a pre-3.59.6 QuickFox build immediately.
Vulnerability watch: a critical cPanel & WHM flaw lets any hosting customer run SQL commands as database root
A critical privilege-escalation flaw in cPanel & WHM, tracked as CVE-2026-58048 (CVSS 9.4), lets an authenticated hosting customer with ordinary MySQL/MariaDB access execute arbitrary SQL commands with full database administrative privileges. The root cause sits in cPanel’s database-rename operation, which fails to preserve the original SQL mode configuration — the result is that renaming a database can cause subsequent SQL to execute in a root context instead of the calling user’s own restricted one. Depending on the underlying OS and database engine configuration, that database-level compromise can extend to operating-system-level access on the shared server. Patches are out in cPanel & WHM 11.110.0.137, 11.118.0.71, 11.126.0.78, 11.134.0.48, 11.136.0.32, and 138.1.6 for WP² deployments. Where patching has to wait, the documented workaround is setting cpsrvd_keepalives_disabled=1 in /var/cpanel/cpanel.config and restarting cpsrvd, which forces a fresh TCP/TLS connection per request on ports 2083, 2087, and 2096 at the cost of extra latency and CPU load. The Hacker News · Cyber Security News
This one matters most in shared hosting: if your infrastructure — or a vendor’s — runs multi-tenant cPanel & WHM, one customer’s account is all it takes to escalate to database root and potentially pivot to other tenants on the same box. Patch the control panel first, not just individual customer accounts, and if you can’t patch today, apply the cpsrvd_keepalives_disabled workaround and watch for it in your next maintenance window rather than leaving the box exposed indefinitely.
Final thought
An AI tool that trusts the network, a VPN vendor’s own build pipeline, a database rename that forgets to check permissions first — three different products, the same failure mode: a convenience path nobody threat-modeled as an attacker’s front door. Patch cadence and default configuration reviews are still the best defense against all three. If you want help figuring out which of your own convenience features double as unguarded entry points, see our how-to guides or book a discovery call.