Artificial intelligence
Before Granting Access: 7 Controls for an AI Agent
When an agent can read data and take action, the business needs permissions, boundaries, approvals and traceability.
Publications
Perspectives that drive better decisions
Ideas to understand the present and build the future of technology.
Security
Critical Langflow vulnerability: what to review
CVE-2026-10561 shows why running Python inside an AI workflow requires isolation, least privilege and timely updates.
Security
GitLab, Node.js, Java and HTTP/3 alerts: what to update
GitLab, Node.js, Java and several HTTP/3 implementations received important alerts. Review the affected areas, fixed versions and priority actions.
Security
GitHub Copilot CLI: how a file could hide commands
CVE-2026-29783 could hide dangerous commands inside apparently safe operations involving repositories, MCP and prompt injection.
Security
Prompt injection in Langroid: from message to server
CVE-2026-54769 scored 10.0: a manipulated instruction could escape Langroid's sandbox and execute commands on the server.
Security
An AI breached Hugging Face: lessons for companies
The Hugging Face incident shows why autonomous agents need strict boundaries, oversight and controls outside the model.