cybr.cx Daily Digest — August 10, 2026

Critical Vulnerabilities

⚠️ Actively exploited — CVE-2026-8037 | Progress LoadMaster | No CVSS listed (KEV)
An unauthenticated command injection vulnerability in Progress LoadMaster allows attackers to execute arbitrary OS commands by sending crafted input to multiple exposed command endpoints. No authentication required. CISA's remediation deadline was today — if you haven't patched or taken LoadMaster off the public internet, treat this as an emergency. Load balancers are high-value targets for initial access.

⚠️ Actively exploited — CVE-2026-63077 | JetBrains TeamCity | No CVSS listed (KEV)
A deserialization of untrusted data vulnerability in JetBrains TeamCity enables unauthenticated remote code execution via the agent polling protocol. CI/CD infrastructure is a persistent target for supply chain attacks — threat actors with RCE on your build server effectively own your pipeline. Patch immediately or isolate the agent communication interface.

⚠️ Actively exploited — CVE-2026-18556 & CVE-2026-18577 | N-able N-central | No CVSS listed (KEV)
Two authentication bypass vulnerabilities affect N-able N-central, the second (CVE-2026-18577) being an incomplete fix for the first. Both allow attackers to bypass authentication via alternate paths and achieve account takeover. N-central is an RMM platform with deep access to managed endpoints — compromise here cascades to every client in a managed service provider's portfolio. The patch-bypass chaining here is particularly concerning.

⚠️ Actively exploited — CVE-2026-34486 | Apache Tomcat | No CVSS listed (KEV)
A missing encryption vulnerability in Apache Tomcat allows attackers to bypass the EncryptInterceptor, and CISA notes it chains with CVE-2025-24813 for increased impact. If you're running Tomcat with EncryptInterceptor enabled and believe that's sufficient protection, it isn't. Review your Tomcat deployment configuration and apply available patches.

⚠️ Actively exploited — CVE-2026-9198 | IBM Langflow | No CVSS listed (KEV)
A code injection flaw in Langflow — the AI workflow orchestration platform — allows unauthenticated attackers to achieve full remote code execution on default deployments. Langflow instances are frequently spun up quickly for prototyping and left exposed. If your organisation is running any AI agent tooling built on Langflow, audit exposure now.

CVE-2026-19341 | UTT HiPER 1200GW (≤2.5.3-170306) | CVSS 8.8 — HIGH
A stack-based buffer overflow in the strcpy call within /goform/pptpSrvGlobalConfig allows remote attackers to corrupt memory via a maliciously crafted EncryptionMode argument. The vendor did not respond to pre-disclosure contact, and a working exploit is public. Small business and branch-office gateway devices with no vendor patch path are the worst kind of problem.

CVE-2026-19346 | Tenda CH22 1.0.0.1 | CVSS 8.8 — HIGH
Command injection in the formCertListInfo function of the Tenda CH22 router allows remote code execution via a manipulated Name parameter. Exploit is public. Tenda consumer-grade hardware appearing in SMB and home networks remains a reliable source of trivially exploitable CVEs — network segmentation is your best mitigation if firmware updates aren't forthcoming.

CVE-2026-19342 / 19343 / 19344 | code-projects Task Management System 1.0 | CVSS 7.3 — HIGH
Three separate vulnerabilities in the same open-source task management application: an authentication bypass via the login password field, SQL injection in the admin login via email/password parameters, and further SQL injection via task_id. All exploits are public. If this codebase is deployed anywhere in your organisation — including internal tooling — take it offline until patched.

CVE-2026-19351 | node-sql-query 0.1.25–0.1.28 | CVSS 7.3 — HIGH
SQL injection in the SelectQuery.from/SelectQuery.build functions of the node-sql-query npm library allows remote exploitation via crafted request parameters. Upgrading to a patched version is the fix. Audit your Node.js dependency trees — this kind of transitive dependency vulnerability is easy to miss and trivial to exploit.

Headline News

OpenAI Pauses Development of AI Model "Astra" Over Autonomous Exploitation Capability
OpenAI has announced it is pausing development work on an AI model internally referred to as Astra after the system demonstrated an unsettling capability: autonomously identifying and exploiting software vulnerabilities, and carrying out cyberattacks without human direction. The pause reflects a growing tension in offensive AI research — the same capabilities that make AI useful for security testing also make it a potent attack platform if misused or misaligned. For practitioners, this is the clearest public acknowledgement yet from a major AI lab that their models have crossed a threshold into genuine offensive utility. The implications for threat modelling are significant: if a well-resourced AI lab's model can autonomously hunt and exploit bugs, similar capabilities in the hands of nation-state actors or well-funded criminal groups must be assumed. Defenders should be evaluating detection strategies for AI-assisted attack patterns, which may move faster and with less predictable tooling signatures than conventional campaigns.

AI-Powered Business Email Compromise Is Scaling Losses to Billions
Business email compromise (BEC) fraud — already a multi-billion dollar annual problem — is accelerating sharply as attackers integrate AI-generated content, deepfake audio, and real-time voice cloning into their social engineering workflows. Where traditional BEC relied on plausible but detectable impersonation, AI-generated voice calls from a "CFO" or "CEO" that match tone, cadence, and phrasing are now operationally viable at scale. Finance teams and executives remain the primary targets, with wire transfer fraud and credential theft the typical objectives. Practitioners should be pushing for out-of-band verbal confirmation protocols that cannot be satisfied by a phone call alone, and training staff to treat any urgent financial instruction — regardless of how convincing the voice sounds — as requiring a secondary verification step through a known-good channel. Detection at the email gateway layer is increasingly insufficient when the primary attack vector has shifted to voice.

Bitcoin's Red Team Finds Critical Vulnerabilities Across Core Repositories Using AI
A volunteer security effort focused on Bitcoin's open-source ecosystem has disclosed more than a dozen vulnerabilities after scanning approximately 150 repositories using an AI-assisted code review platform they are now open-sourcing. The group — operating as an informal red team — is explicitly targeting critical infrastructure in the Bitcoin stack, including wallet implementations, libraries, and protocol tooling. Several of the disclosed issues were rated critical severity. This matters beyond the cryptocurrency context: the model of AI-augmented volunteer red teams performing continuous scanning of open-source infrastructure is one that the broader security community should be watching closely and potentially replicating for other critical open-source projects. It also demonstrates that the same AI tooling OpenAI is pausing for being too capable is, in constructive hands, finding real bugs in widely deployed code.

Schrödinger's Feed

Scientists have successfully generated quantum entanglement directly from sunlight in an outdoor experiment, producing entangled photons with approximately 94% fidelity to an ideal state — without the high-power lasers typically required. This is a materials and optics breakthrough as much as a quantum one, and it hints at future quantum key distribution (QKD) networks that could operate from ambient light rather than controlled lab conditions. The energy and infrastructure implications are substantial: cheaper, more deployable QKD could accelerate the transition to quantum-secured communications well ahead of current timelines. Practitioners planning post-quantum cryptography migration roadmaps should note that the "harvest now, decrypt later" threat window may be shorter than assumed if QKD deployment costs drop significantly.

/dev/random

In 2011, someone made a Long Bet — a formal, financially-staked prediction — that a specific URL would still be accessible in 2026. That URL is now dead, making this one of the most elegantly self-defeating predictions in internet history: the bet's evidence page was itself a victim of link rot. The original post predicting the URL's survival has, of course, also disappeared from its original location. It's a perfect closed loop — a monument to the impermanence of the web, hosted on a platform designed specifically to preserve long-term predictions, pointing to a page that no longer exists. The internet: where nothing is permanent except screenshots and Wayback Machine crawls.