Cyber Brief: NCSC AI guidance, Entra ID flaw, Rust supply chain attack

Today's brief focuses on the practical security challenges emerging from increasingly autonomous technology and the importance of maintaining clear oversight across identity systems, software supply chains and enterprise infrastructure. The NCSC has published new guidance on managing agentic AI systems, Microsoft has patched a critical Entra ID vulnerability already exploited in the wild, a supply chain attack has compromised widely used Rust software packages, and Cisco has released fixes for multiple critical vulnerabilities in its Secure Workload Software. Each of these developments highlights the need for organisations to maintain active visibility, structured governance and proportionate controls across systems that are becoming more complex and interconnected.

NCSC publishes guidance on managing cyber risk in agentic AI systems

The National Cyber Security Centre has published new guidance on managing the cyber security risks associated with agentic AI systems, which are autonomous systems capable of making decisions and taking actions without direct human intervention. The NCSC blog post, published on 20 August 2026, explains that while these systems offer significant operational benefits, they also introduce new risks through unintended activity, privilege escalation and the potential for adversarial manipulation. The guidance recommends that organisations implement safeguards including sandboxing, active oversight, strict access controls and clear logging to ensure that autonomous actions remain within acceptable boundaries.

For UK businesses beginning to deploy or experiment with agentic AI tools, this guidance provides a practical framework for managing systems that can act independently within enterprise environments. The operational risk is that autonomous agents, if not properly constrained, may access sensitive data, modify configurations, execute commands or interact with external systems in ways that were not anticipated or authorised. This is particularly relevant for organisations using AI-powered automation in cloud environments, DevOps pipelines, customer service platforms or internal workflow tools. The NCSC's emphasis on sandboxing and oversight reflects the reality that these systems require a different approach to risk management compared to traditional software.

Why it matters

For UK businesses exploring or deploying agentic AI, this is a prompt to review how autonomous systems are being scoped, tested and monitored within your environment. Ensure that AI agents operate within clearly defined boundaries, that their actions are logged and reviewable, and that access to sensitive systems or data is tightly controlled. This guidance helps organisations realise the benefits of automation while maintaining the visibility and control necessary to manage unintended consequences.

Source: NCSC UK

Microsoft patches critical Entra ID vulnerability exploited in the wild

Microsoft has issued an urgent security update for a maximum-severity vulnerability in Microsoft Entra ID, the company's cloud-based identity and access management service, which was previously known as Azure Active Directory. The vulnerability, tracked as CVE-2026-69836 and assigned a CVSS score of 10.0, is a remote code execution flaw that Microsoft confirmed has been exploited in the wild. The Hacker News reported on 21 August 2026 that Microsoft has stated no customer action is required, indicating that the fix has been applied automatically across the service. The vulnerability represents a critical risk to identity infrastructure, which underpins access control across cloud services, applications and enterprise resources.

For UK organisations relying on Microsoft Entra ID for identity and access management, this incident underscores the operational importance of cloud-based identity services and the potential impact when vulnerabilities in these systems are exploited. Entra ID is widely used to manage user authentication, conditional access policies, multi-factor authentication and integration with third-party applications. A remote code execution vulnerability in this service could allow an attacker to compromise identity infrastructure, bypass access controls or gain elevated privileges across connected systems. While Microsoft has applied the fix automatically, organisations should review recent authentication logs, conditional access policies and any unusual sign-in activity to ensure no unauthorised access occurred during the window of exploitation.

Why it matters

For many organisations, identity infrastructure is the foundation of access control across cloud and on-premises environments. This is a prompt to review recent sign-in logs, audit trails and conditional access policy changes within your Entra ID tenant. Ensure that monitoring and alerting for unusual authentication activity is in place and that your identity governance processes include regular reviews of privileged accounts, service principals and application permissions.

Source: The Hacker News

Rust supply chain attack compromises packages with 245 million downloads

The Rust Project has removed malicious versions of three widely used Rust software packages from the crates.io repository after a compromised maintainer account was used to publish releases containing a typosquatted dependency. The Hacker News reported on 21 August 2026 that the affected packages are arrayref version 0.3.10, internment version 0.8.7, and append-only-vec version 0.1.9, which collectively have been downloaded approximately 245 million times. The malicious releases included a build script that downloaded and executed a remote payload during the compilation process, meaning that any developer or build system compiling software using these versions would have executed the attacker's code. The Rust Project has confirmed that the malicious versions have been deleted and that the maintainer account has been secured.

For UK organisations using Rust in software development, DevOps pipelines or infrastructure tooling, this incident highlights the operational risk of supply chain attacks targeting widely used open source dependencies. The attack method, which executed malicious code during the build process rather than at runtime, means that traditional application security testing may not have detected the compromise. Any organisation that compiled software using the affected package versions during the window of compromise should treat their build environments, CI/CD pipelines and any artefacts produced during that period as potentially compromised. This includes reviewing build logs, checking for unexpected network connections during compilation, and assessing whether any credentials, secrets or sensitive data were accessible to the build environment.

Why it matters

For organisations using Rust or managing software supply chains, this is a prompt to review your dependency management practices, build environment security and software composition analysis processes. Ensure that build systems operate in isolated environments with limited access to production credentials, that dependency updates are reviewed before being merged, and that you maintain visibility over what code is being executed during the build process. Consider implementing additional controls such as dependency pinning, checksum verification and build reproducibility checks.

Source: The Hacker News

Cisco releases fixes for critical vulnerabilities in Secure Workload Software

Cisco has published security advisories addressing five critical and high-severity vulnerabilities in its Secure Workload Software, a platform used for application dependency mapping, micro-segmentation and workload protection in data centre and cloud environments. The Register reported on 21 August 2026 that the vulnerabilities have been assigned CVSS scores of 10.0, 10.0, 9.9, 9.6 and 7.5, indicating a range of serious security issues. The flaws affect both on-premises and SaaS deployments of the software, and Cisco has confirmed that updates are available for all affected versions. Organisations using Cisco Secure Workload Software are advised to apply the patches as a priority, as the vulnerabilities could allow attackers to execute arbitrary code, escalate privileges or bypass security controls.

For UK businesses using Cisco Secure Workload Software to manage application security, network segmentation or workload visibility, these vulnerabilities represent a significant risk to infrastructure that is often deployed to enforce security policies across critical systems. Secure Workload Software typically has privileged access to network traffic, application metadata and security policy enforcement mechanisms, meaning that a compromise could allow an attacker to disable protections, intercept sensitive data or move laterally across segmented environments. Even organisations using the SaaS version of the platform are required to apply updates, indicating that some components may be customer-managed or that configuration changes are necessary to fully address the vulnerabilities.

Why it matters

For organisations using Cisco Secure Workload Software, this is a prompt to review your patch management process and ensure that updates are applied to both on-premises and SaaS deployments. Verify that your security monitoring includes visibility over changes to workload protection policies, segmentation rules and access controls, and ensure that any anomalies during the patching window are investigated. This is also an opportunity to review whether your security tooling itself is being monitored and maintained with the same rigour as the systems it protects.

Source: The Register

Today's Key Actions

• Review how agentic AI systems are being deployed, tested and monitored within your organisation, ensuring that autonomous actions are sandboxed, logged and subject to clear access controls aligned with the NCSC's published guidance.
• Check recent Microsoft Entra ID sign-in logs, audit trails and conditional access policy changes for any unusual activity, and ensure that monitoring and alerting for identity-related anomalies is in place and actively reviewed.
• Assess whether your organisation uses Rust in software development or build pipelines, and review dependency management practices, build environment security and software composition analysis processes to reduce supply chain risk.
• Apply available patches for Cisco Secure Workload Software across both on-premises and SaaS deployments, and verify that security monitoring includes visibility over changes to workload protection policies and segmentation rules.
• Ensure that ownership and accountability for identity infrastructure, software supply chain security, security tooling and autonomous systems is clearly defined and that review processes are in place across these areas.

Secarma Insight

The common thread across today's stories is the importance of maintaining active oversight and structured governance across systems that are becoming more autonomous, interconnected and foundational to how organisations operate. Whether it's identity infrastructure, software supply chains, security tooling or emerging AI capabilities, the organisations that manage these areas well are those that have already established clear ownership, regular review processes and proportionate controls before incidents occur. Good security practice is not about reacting to every new vulnerability or threat, but about building the habits, visibility and discipline that allow you to respond confidently and effectively when issues arise. The goal is to feel informed and prepared, not perpetually reactive.