cybr.cx — Daily Digest | October 07, 2026

Critical Vulnerabilities

⚠️ Actively exploited — CVE-2026-88779 | Citrix NetScaler ADC / Gateway
CISA KEV listed (due date: today). An out-of-bounds memory buffer vulnerability in Citrix NetScaler ADC and NetScaler Gateway enables denial-of-service conditions. Exploitation is confirmed in the wild — if you haven't patched or mitigated this already, you're overdue. Citrix gateway infrastructure is perennially high-value for threat actors; treat this as urgent.

⚠️ Actively exploited — CVE-2026-102489 + CVE-2026-102490 | Zammad (Zammad GmbH)
CISA KEV listed. These two flaws chain together for a particularly ugly outcome: CVE-2026-102489 is a session fixation bug that enables remote code execution as the zammad user, and CVE-2026-102490 is a privilege escalation that takes that foothold all the way to root. Organisations running Zammad for ticketing and customer support — often with broad internal access — should patch both simultaneously. CISA's remediation deadline has already passed.

⚠️ Actively exploited — CVE-2026-104286 | Fortinet FortiMail
CISA KEV listed. A path traversal combined with improper null byte handling lets an unauthenticated attacker write arbitrary files to the underlying system via crafted HTTP/HTTPS requests — effectively an unauthenticated file write to any path the process can reach. FortiMail sits on the network perimeter handling inbound email, making it an attractive entry point. Patch or isolate immediately.

⚠️ Actively exploited — CVE-2026-76504 | Cisco Catalyst SD-WAN Manager
CISA KEV listed. Improper URI hex-encoding handling lets an unauthenticated remote attacker access the SD-WAN Manager with full admin privileges. SD-WAN management planes typically have sweeping visibility across branch infrastructure — a compromise here can cascade broadly. CISA's deadline has passed; prioritise accordingly.

⚠️ Actively exploited — CVE-2026-86950 | Apple iOS, macOS, iPadOS
CISA KEV listed. An out-of-bounds write in CoreGraphics can lead to arbitrary code execution across Apple's major platforms. Given the ubiquity of Apple devices in enterprise environments and the confirmed in-the-wild exploitation, ensure devices are updated and MDM compliance policies are enforced.

CVE-2026-106038 | Mooncake Store ≤ 0.3.13.post1 | CVSS 8.2
A missing authentication flaw on the coro_rpc port allows unauthenticated attackers to force-delete any cached object — including wiping entire stores via RemoveAll or regex-matched batch deletion. Distributed systems relying on Mooncake for caching could face cascading request failures or worse if adversaries use this to selectively poison cache state. Update immediately.

CVE-2026-77226 | Camunda 7.24.0–7.24.14 | CVSS 8.1
The Admin web application's first-run setup endpoint miscounts administrators — it only checks direct group members of camunda-admin, not all configured admins — allowing an unauthenticated attacker to invoke the setup endpoint and create a new admin user on a live instance. Any internet-exposed Camunda instance on an affected version is at serious risk of complete takeover. Upgrade to 7.24.15.

CVE-2026-82531 | Smarty < 4.5.8 and 5.x < 5.8.5 | CVSS 8.1
A code injection flaw in Smarty's template inheritance system leaves nocache_hash null during extends: processing, allowing an attacker who can supply assigned template data to inject a forged SmartyNocache marker into the PHP cache file. The result is arbitrary PHP execution on next cache include. Smarty is embedded in a large number of PHP applications and CMSes — check your dependency tree.

CVE-2026-104073 | NetBox 2.9.5–4.6.x | CVSS 7.6
A server-side template injection in NetBox's custom links feature exposes the raw Django HttpRequest object to Jinja2, letting any low-privileged user with "Can add custom links" permission steal session cookies and API tokens from other users — including admins. NetBox is widely used for network source-of-truth; an API token harvest here can mean full network inventory exposure. Upgrade to 4.7.0.

CVE-2026-82531 / CVE-2026-105840 + CVE-2026-105841 | lrzsz < 0.13.0 | CVSS 7.5 each
Two separate flaws in the lrz ZMODEM receive utility: one allows path traversal via absolute filenames to overwrite arbitrary writable files; the other is an OS command injection in pipe mode where unescaped filenames containing shell metacharacters are passed directly to popen(). Both are exploitable by a malicious sender during a file transfer. Legacy serial/modem tooling is still found in embedded systems and industrial environments — worth auditing.

CVE-2026-105835 | PLANKA 2.2.0–2.2.1 | CVSS 7.4
No rate limiting on TOTP verification means an attacker with a valid password can reuse the 10-minute pending session token to brute-force a six-digit 2FA code — roughly 1 million guesses in a window where the expected search space is a few hundred thousand. MFA is meant to be the last line of defence; this effectively nullifies it. Update PLANKA or block the endpoint at the perimeter.

Headline News

Meta's Muse AI Agent Launched with Known VM Escape Vulnerability

In the weeks immediately preceding the public launch of Meta's Muse AI agent product, engineers internally identified several security vulnerabilities — including at least one VM escape flaw that could have allowed malicious users to break out of the sandboxed execution environment. The vulnerability was reportedly fixed before launch, but the timeline raises serious questions about the adequacy of pre-release security review cycles for AI agent products, where the attack surface is novel and poorly understood by traditional AppSec tooling. Beyond the VM escape, published analysis identifies additional privacy and security concerns around how Muse handles user data and what access the agent retains during and after sessions. For practitioners, this serves as a case study in the unique risks of agentic AI systems: unlike traditional web apps, agents may have privileged access to host resources, long-lived sessions, and user context that dramatically raises the stakes of any sandbox failure. Security teams evaluating AI agent deployments should be demanding documented threat models and third-party security assessments — not press releases.

Rejetto HFS Actively Scanned for RCE Flaw CVE-2026-61500

Mass scanning activity has been detected targeting Rejetto HTTP File Server (HFS) instances, with attackers probing for CVE-2026-61500 — a weak signing key vulnerability that enables session forgery, account takeover, and full remote code execution. HFS is lightweight file-sharing software popular in small business and home lab environments, meaning many installations are unmanaged and unlikely to be on active patch cycles. The weak signing key means an attacker who identifies the key can forge authenticated sessions without needing any credentials, making exploitation trivially scriptable once the key is known. Active scanning campaigns like this typically precede broader exploitation waves within days — operators running any version of Rejetto HFS should take the service offline or apply mitigations immediately. Shodan and Censys enumeration likely gives attackers a ready list of targets.

ASOS Push Notifications Hijacked in Apparent Credential or API Compromise

Customers of the UK-based fashion retailer ASOS received unsolicited push notifications apparently sent by third parties rather than ASOS itself, pointing to a compromise of push notification infrastructure, API credentials, or an associated third-party service. While the content of the notifications was not reported as overtly malicious, the incident illustrates how push notification channels — often deprioritised in security reviews — can be weaponised for phishing, fraud, or social engineering at scale against a trusted brand's entire user base. For practitioners, this is a reminder that mobile push notification pipelines involve third-party SDKs, cloud messaging services, and API keys that need the same credential hygiene and access controls as any other production system. A threat actor with access to push infrastructure can craft convincing in-app messages that users are highly conditioned to trust.

Schrödinger's Feed

Florida Atlantic University has received a $200,000 NIST grant to build a quantum cybersecurity workforce consortium in South Florida, partnering with D-Wave Quantum to integrate quantum computing, cybersecurity, and critical energy infrastructure protection into hands-on training programmes. This is part of a broader recognition that the post-quantum transition isn't just a cryptography problem — it's a workforce problem, and the gap between NIST finalising PQC standards and organisations actually deploying them is largely a skills gap. The grant reflects institutional acknowledgement that defending critical infrastructure against quantum-capable adversaries requires practitioners who understand both domains fluently, not just theoretically. Security teams should note: the window between "standards exist" and "your organisation is actually PQC-compliant" is closing faster than hiring pipelines can fill it.

/dev/random

The tapo open-source library — a Rust/Python client for TP-Link smart home devices — has added support for TP-Link's TPAP protocol, the newer communication layer used in more recent Kasa and Tapo hardware. What makes this genuinely interesting from a security lens is that community reverse engineering of undocumented vendor protocols consistently reveals how these devices actually communicate, often turning up hardcoded keys, weak authentication, or cleartext fallbacks that vendor documentation politely omits. This isn't the first time the tapo project surfaced protocol quirks that the official SDK glossed over. If you've got TP-Link devices on your network — and statistically, you do — it might be worth a look at what the open-source community has learned about how they actually behave.