What does BroadVision cover in its News & Insights?

Practical analysis and field reports from active engagements, managed IT, networks, cloud migration, cybersecurity, data and AI, drawn from BroadVision's work worldwide and 47 countries.

How often does BroadVision publish news?

We publish in-depth posts weekly, plus shorter project notes whenever a meaningful deployment, threat advisory or vendor change warrants it.

Where is BroadVision based?

BroadVision HUB is in Mulberry Hill Office Park, 4 Broadacres Drive, Dainfern, Johannesburg. Our teams operate worldwide and 47 countries.

The IT pulse, every Friday.

A weekly briefing on the developments shaping enterprise IT, cloud, AI, cybersecurity and SaaS, sourced from reputable publishers across the past seven days, interpreted by BiVi Intelligence for IT leaders, with every story linked back to its original report. The previous fortnight stays available below.

Cybersecurity

Git Configuration Flaws Enable Remote Code Execution in AI Coding Agents

2 September 2026

Manifold Security has identified eight vulnerabilities in AI coding tools that allow malicious Git configurations to execute unauthorized commands on developer machines.

Git Configuration Flaws Enable Remote Code Execution in AI Coding Agents

Security researchers at Manifold Security have disclosed eight vulnerabilities affecting seven popular command-line AI coding agents, including Claude, Codex, and Cursor. The flaws stem from how these agents interact with a repository's Git configuration. If a repository contains a malicious configuration naming a specific command, the AI agent may execute that command on the developer's local machine without requesting approval or alerting the user.

The exploitation occurs because the commands are executed as the local user and operate outside of the AI agent's sandbox environment. At the time of publication, four of these security flaws remained unpatched. To trigger the exploit, a developer must simply open or interact with a compromised repository using a vulnerable AI agent, allowing the attacker-defined code to run with the developer's permissions.

This discovery highlights a significant attack vector targeting software supply chains and developer environments. By embedding malicious instructions within standard configuration files, attackers can bypass traditional sandboxing measures intended to isolate AI processes. The risk is particularly high for teams that frequently clone external repositories or use third-party code for development tasks.

IT leaders and operations managers should verify the patch status of AI tools used within their development teams and consider implementing stricter policies regarding the use of AI agents with untrusted repositories. Monitoring for unusual local command execution originating from development tools can help mitigate the risk of unauthorized access or data exfiltration.

The BroadVision view

These vulnerabilities demonstrate that even advanced AI productivity tools can introduce unforeseen risks into the software development lifecycle. Mid-market IT teams must balance the speed of AI adoption with rigorous governance over developer environments and repository access. Implementing comprehensive managed IT services can help organizations maintain visibility and security over evolving developer toolchains. https://broadvision.com/managed-it-services Teams weighing what to change first can review BroadVision's managed IT services.

Have a story for us, or a problem we can write about?

Get in touch