cybr.cx Daily Digest — August 06, 2026

Critical Vulnerabilities

⚠️ Actively exploited — CVE-2026-9198 | IBM Langflow OSS | CVSS: Critical
CISA added this to the KEV catalogue on August 4th with a patch deadline already passed. Unauthenticated attackers can achieve full remote code execution on default Langflow deployments via a code injection flaw — no credentials required. If you're running any Langflow instance, treat this as a fire drill: isolate, patch, or take offline immediately. This one is being actively weaponised in the wild.

⚠️ Actively exploited — CVE-2026-63077 | JetBrains TeamCity | CVSS: Not yet scored (NVD)
Added to CISA KEV on August 5th with a patch due August 8th — that's tomorrow. The flaw is a deserialization of untrusted data vulnerability in the agent polling protocol, enabling unauthenticated remote code execution. TeamCity's central role in CI/CD pipelines makes this extremely high-value for supply chain attackers. Patch now; don't wait for the deadline.

⚠️ Actively exploited — CVE-2026-18556 & CVE-2026-18577 | N-able N-central | CVSS: Not yet scored (NVD)
Two authentication bypass vulnerabilities in N-able N-central, both actively exploited. CVE-2026-18577 is the more alarming of the two — it's an incomplete patch for CVE-2026-18556, meaning attackers pivoted around the first fix almost immediately. Managed service providers using N-central for endpoint management should treat both as critical-priority patches. Full account takeover is the confirmed impact.

⚠️ Actively exploited — CVE-2026-34486 | Apache Tomcat | CVSS: Not yet scored (NVD)
Attackers can bypass Tomcat's EncryptInterceptor due to missing encryption of sensitive data in cluster communication. CISA's patch deadline was August 7th. Tomcat's ubiquity across enterprise Java deployments makes this a broad-surface risk; verify your cluster configurations and confirm the patch is applied across all nodes.

⚠️ Actively exploited — CVE-2026-20316 | Cisco Secure Firewall Management Center (FMC) | CVSS: Not yet scored (NVD)
A hard-coded password embedded in Cisco FMC allows unauthenticated remote login via a low-privileged account. CISA's remediation deadline has already passed (August 1st). If you haven't patched this yet, assume exposure — hard-coded credentials are trivially discovered once publicly documented.

CVE-2026-10025 | IBM QRadar 7.6.0.0–7.6.0.1, 7.5.0 UP 15 IF 005 | CVSS: 8.2
An XXE injection vulnerability in QRadar's parseXmlPayload() function within q1labs_core.jar can be triggered by sending crafted XML-formatted syslog events when XML-format property autodetection is enabled on any log source. An attacker with log injection capability could read local files or initiate server-side request forgery. IBM has issued fixes — apply them and review your log source configurations.

CVE-2026-9196 | IBM Langflow OSS 1.0.0–1.10.3 | CVSS: 8.1
Authenticated attackers can abuse the Agentic Assistant validation flow, which executes LLM-generated Python code in the backend before user approval. This allows side effects including outbound network access and filesystem operations triggered via crafted model responses. Combined with CVE-2026-9198 above, Langflow's attack surface right now is substantial — assume no Langflow version in this range is safe.

CVE-2026-70617 | Spacebar Server (pre-commit dcfd910) | CVSS: 8.1
Any authenticated user can add themselves to arbitrary private group DM channels by sending an unauthenticated PUT request to the /channels/{channel_id}/recipients/{user_id} endpoint. No membership verification is performed. Self-hosted Spacebar (the open-source Discord-compatible server) deployments should update immediately; complete message history of private channels is exposed.

CVE-2026-69111 | Milvus through 2.6.22 and 3.0.0 | CVSS: 7.5
An unauthenticated attacker can terminate Milvus service components by hitting the unprotected /management/stop endpoint on port 9091 with a crafted GET request. The endpoint bypasses REST API authentication middleware entirely. Organisations using Milvus for vector database workloads — increasingly common in AI pipelines — should firewall port 9091 from untrusted networks immediately, regardless of patch status.

Headline News

SharePoint Exploits Breach Switzerland's Federal IT Agency

Switzerland's Federal Office for Information Technology and Telecommunications (FOITT) has confirmed that attackers exploited vulnerabilities in Microsoft SharePoint to compromise approximately 200 staff accounts across the federal IT infrastructure. The breach is significant not only in scope but in target: FOITT manages IT services for numerous Swiss federal agencies, meaning lateral access from a single SharePoint compromise could have cascading effects across government operations. Investigators are still determining the full extent of data exfiltration, and affected servers are actively being rebuilt — a sign the incident was severe enough to warrant full remediation rather than in-place recovery. For practitioners, this reinforces that SharePoint, despite its enterprise ubiquity, remains a high-value entry point; unpatched CVEs and misconfigured external sharing settings are routinely weaponised against government and enterprise targets alike. Audit your SharePoint exposure now.

WebKit IP and DNS Leaks Undermine Proxy Browsers and iCloud Private Relay

A disclosed vulnerability in WebKit is causing real IP and DNS leaks in browsers and privacy tools that depend on the engine — including iCloud Private Relay users and proxy-reliant browsers on Apple platforms. The flaw means that under certain conditions, WebKit initiates connections that bypass the configured proxy or relay tunnel, sending identifiable traffic directly to external hosts. This is particularly significant for threat analysts, journalists, and anyone relying on iCloud Private Relay or VPN-adjacent browser-level privacy tools on iOS and macOS. The root issue lies in how WebKit handles specific resource loads that circumvent the system's network abstraction layer. Until a patched WebKit ships in a forthcoming OS update, users with strong anonymity requirements should treat Apple's built-in privacy tools as unreliable and consider alternative mitigations at the network layer.

Coldcard Hardware Wallet Firmware Bug Poisoned Key Generation Entropy

A firmware vulnerability in the Coldcard hardware Bitcoin wallet was found to critically weaken entropy during cryptographic key generation, potentially producing private keys with far lower randomness than expected. The flaw, present across affected firmware versions, means wallets generated during the vulnerable window may be susceptible to brute-force recovery — a particularly severe outcome for a device whose entire value proposition is air-gapped, offline key security. Researchers describe the bug as a "ticking time bomb": keys already generated remain at risk even after firmware updates, because the vulnerability affects the material created at wallet setup time, not ongoing operations. Users who set up a Coldcard wallet on a vulnerable firmware version should treat any funds in those wallets as potentially exposed and migrate to freshly generated wallets on patched firmware as a precaution. This is a sobering reminder that hardware security modules are not immune to software-layer vulnerabilities — and that entropy bugs are among the most catastrophic in cryptographic engineering.

Schrödinger's Feed

Researchers have announced a quantum encryption method they claim cannot be physically copied — a property rooted in the quantum no-cloning theorem applied to a new key distribution scheme. Unlike classical cryptographic assumptions (which rely on computational hardness), this approach makes interception structurally detectable rather than merely difficult: any attempt to duplicate the quantum state irreversibly disturbs it. The practical implementation details are still emerging, but the underlying claim, if it holds up to peer scrutiny, points toward a class of communications channels resistant to "harvest now, decrypt later" attacks entirely. Practitioners building long-lifecycle data protection strategies — think government archives, healthcare records, or financial instruments with decade-plus sensitivity windows — should track this research closely as post-quantum standards mature and hardware implementations inch toward production viability.

/dev/random

The security community's collective attention briefly drifted from CVEs this week when Google announced that Demis Hassabis is stepping up to Chair of DeepMind while Jeff Dean — one of the architects of modern machine learning infrastructure — departs the company entirely. The real curiosity for security practitioners: as the people who built the foundational AI systems underpinning half the industry's threat detection tooling move on, questions about institutional knowledge transfer become surprisingly relevant. It's not quite a vulnerability, but "key person dependency" is a legitimate risk category, and few organisations have more key persons than DeepMind. No CVE has been assigned yet, though some would argue the CVSS scoring rubric needs a "vibes" metric.