Posts

🟠 High | Source: Microsoft Security Response Center CVE-2026-14647 is an out-of-bounds memory access vulnerability in ONNX Runtime, Microsoft’s open-source machine learning inference engine, specifically within the convPoolShapeInference_opset19 function in old.cc. This type of memory corruption flaw can potentially be exploited to crash an application or, in more severe scenarios, execute arbitrary code. It is particularly relevant to organisations using ONNX Runtime for AI/ML inference workloads on Azure or in containerised cloud environments. ...

🟠 High | Source: Microsoft Security Response Center A vulnerability in the Keras deep learning library (CVE-2026-12480) allows an attacker to bypass virtual dataset protections in the HDF5 file format, enabling arbitrary file reads on the host system. This could expose sensitive data, configuration files, or credentials on systems where Keras is used to load untrusted model files. It is particularly relevant to cloud-based ML training and inference environments where user-supplied or third-party model files are processed. ...

🟠 High | Source: Microsoft Security Response Center CVE-2026-54891 is a TLS vulnerability in which plaintext APPLICATION_DATA is injected during the TLS handshake and subsequently delivered to the client application after the handshake completes. This means data intended to be protected by encryption may be processed by the application before a secure channel is fully established, potentially exposing sensitive information or enabling data manipulation. The flaw is particularly concerning in cloud environments where TLS is the primary transport security mechanism. ...

🟠 High | Source: Microsoft Security Response Center A vulnerability in SSH SFTP server implementations allows an attacker to trigger an infinite loop by sending specially crafted extended channel data, causing a denial of service. This affects Azure-hosted or Microsoft-managed services relying on the vulnerable SSH/SFTP component. If exploited, legitimate users and automated processes depending on SFTP connectivity could be completely locked out. Security Architect’s Take: Review any Azure services or custom workloads exposing SFTP endpoints and apply the relevant Microsoft patch immediately. In the interim, consider restricting SFTP access via network security groups or Azure Firewall rules to trusted IP ranges only, reducing the attack surface until patching is confirmed. ...

🟠 High | Source: Microsoft Security Response Center CVE-2026-55952 is a denial of service vulnerability in TLS 1.3 that can be triggered by sending a malformed ClientHello message containing a crafted pre-shared key (PSK) extension. An attacker can exploit this remotely to crash or make unavailable services relying on the affected TLS implementation. This is particularly significant in cloud environments where TLS termination is widespread and availability is business-critical. Security Architect’s Take: Review which Azure services and self-managed workloads use the affected TLS 1.3 stack and prioritise patching, particularly for internet-facing endpoints and API gateways. In the interim, consider whether Web Application Firewall (WAF) rules or network-layer controls can block or inspect malformed ClientHello messages as a temporary mitigation. ...

🟠 High | Source: AWS Security Bulletins A SQL injection vulnerability (CVE-2026-14471) has been identified in AWS’s open-source mcp-gateway-registry, specifically in the metrics-service retention policy component. An authenticated attacker can supply a malicious table_name value to execute arbitrary SQL against the metrics database, potentially exposing API keys and allowing data manipulation or deletion. Versions 1.0.3 through 1.0.12 are affected. Security Architect’s Take: Update mcp-gateway-registry to a version beyond 1.0.12 immediately, and in the interim audit who has authenticated access to the metrics-service retention policy endpoint — treat any stored API keys as potentially compromised and rotate them as a precaution. ...

🟠 High | Source: The Hacker News An Iranian state-linked hacking group, associated with Iran’s Ministry of Intelligence and Security, has deployed a newly discovered modular command-and-control framework called Cavern against Israeli IT providers and government organisations. The framework is previously undocumented, suggesting active development of new offensive tooling by this threat actor. The targeting of IT providers raises particular concern due to the potential for downstream supply chain compromise. Security Architect’s Take: Review egress traffic and C2 detection rules within your SIEM for anomalous outbound connections, particularly from systems managed by third-party IT providers. If your organisation operates in sectors targeted by MOIS-affiliated groups — including government supply chains — consider increasing threat hunting cadence and validating that endpoint detection tooling has up-to-date signatures for novel C2 frameworks. ...

🟠 High | Source: The Register — Security A member of the European Parliament investigating spyware has had their phone infected with Pegasus, the notorious commercial surveillance tool developed by NSO Group. This comes as campaigners push the EU to act on long-stalled recommendations from the PEGA Committee, which concluded its inquiry into member state misuse of spyware in 2023. The incident underscores that even those scrutinising spyware are not immune, raising serious concerns about the security of EU institutional devices and political processes. ...

🟠 High | Source: AWS Security Blog When AI agents delegate tasks to other agents in a chain, the authorisation scope can silently expand beyond what the original user permitted — a problem known as privilege escalation in agentic systems. This AWS blog post demonstrates how to use Cedar, AWS’s open-source policy language, to enforce least-privilege controls at every hop in a multi-agent chain. This matters because RBAC alone is insufficient to contain authorisation drift across complex, autonomous AI workflows. ...

🟠 High | Source: The Hacker News A suspected Chinese threat actor is targeting Indian taxpayers, tax professionals, and corporate finance teams using spear-phishing emails that impersonate India’s Income Tax Department. Victims are lured into running a fake tax filing utility that ultimately installs DcRAT, a remote access trojan capable of stealing sensitive data. The campaign, dubbed Operation DragonReturn by Seqrite Labs, is a multi-stage attack with clear espionage and data exfiltration objectives. ...