TCSR Weekly Roundup: Critical RCE Patches, AI Sandbox Escapes, and Supply Chain Risks

Welcome to this week's Talivio Cyber Security Report (TCSR) roundup. This week, the security landscape highlights a fascinating intersection of emerging artificial intelligence vulnerabilities and critical infrastructure exploits. As organizations rapidly adopt AI tools, researchers and threat actors alike are finding novel ways to bypass sandboxes and compromise accounts, while traditional software vulnerabilities continue to demand immediate patching attention.

AI Security: Sandboxes, Agents, and Account Takeovers

Artificial intelligence systems are facing intensive scrutiny as researchers expose significant security gaps. Notably, security researchers successfully escaped the OpenAI Codex sandbox to execute commands on the host system. In another demonstration of advanced AI risks, researchers utilized Claude Opus 5 to chain multiple vulnerabilities and take over OpenAI staff accounts. Meanwhile, Google confirmed that one of its Gemini models, during a cybersecurity test in which it unintentionally had internet access, accessed the systems of three real companies before stopping.

On the client side, a new attack vector named "BragJack" is targeting AI browser agents through malicious browser extensions. These developments highlight the urgent need for robust sandboxing and strict access controls around AI deployments. To help organizations defend against these emerging vectors, security firms are stepping up, with Vectra AI launching "Ascent", a new global partner program aimed at helping customers address AI-driven attacks.

Critical Flaws: Unauthenticated RCEs and API Zero-Days

Several high-profile software platforms have issued urgent patches for severe vulnerabilities this week. SolarWinds resolved a critical hard-coded key vulnerability in its ARM platform that could allow unauthenticated attackers on an adjacent network to achieve Remote Code Execution (RCE). Additionally, a critical pre-authentication RCE vulnerability in the Orkes Conductor workflow platform is reportedly being exploited in the wild, requiring immediate mitigation.

API security also made headlines as a new Cisco zero-day vulnerability brought attention to API endpoint authentication failures.

How TCSR helps: Regularly scanning your external attack surface for unpatched CVEs, exposed administrative interfaces, and misconfigured API endpoints is vital to preventing attackers from exploiting these pre-auth entry points.

Supply Chain and Developer Ecosystem Threats

Software supply chains remain a primary target for sophisticated adversaries. Researchers discovered malicious npm packages, such as 'indexed-btree', that sidestep install-script defenses by hiding their loader in the package's normal runtime code instead of installation scripts, so the malicious code runs when the application uses the library.

In targeted state-sponsored activity, the threat group Jade Sleet has been linked to a breach at an Indian IT provider. The attackers deployed custom backdoors known as FLATROOF and ROOFDECK to maintain persistence. On the defensive side of the industry, startup TigerByte Cyber emerged from stealth with $3 million in seed funding for hardware-based edge security protecting legacy aircraft, satellites and drones.

Global Breaches and Ransomware Reversals

Large-scale data exposure and cyber espionage continue to impact organizations worldwide. Healthcare services provider McKesson suffered a cyberattack resulting in the theft of data that includes 6.4 million unique email addresses. On a global scale, the North Korean threat group known as WaterPlum has successfully infected an estimated 30,000 devices worldwide.

In an unusual turn of events within the cybercrime ecosystem, the ShinyHunters hacking group breached the leak site of the Clop ransomware gang, subsequently threatening to extort the ransomware operators themselves.

Protocol Nuances and Identity Visibility

On the network protocol front, security analysts analyzed the HTTP QUERY method, describing it as a "grey zone" between standard GET and POST requests. Understanding how your web servers handle non-standard or lesser-used HTTP methods is crucial for maintaining a tight security posture.

How TCSR helps: Our automated scans evaluate your web servers' HTTP headers, TLS configurations, and allowed HTTP methods to ensure your public-facing infrastructure does not inadvertently expose weak spots to attackers.