CyberSG Connect: Trustworthy and Secure AI Agents
Kush Kanwar joined CyberSG Connect at SGInnovate for conversations on the governance, security and privacy of AI agents.
Written by
Kush Kanwar
Singapore
Aug 21, 2026
Trustworthy agents need clear limits on the information they use and the actions they can take. Those limits become harder to maintain when an agent connects to another team, system or organisation.
On 21 August 2026, Kush Kanwar joined CyberSG Connect: Trustworthy and Secure AI Agents at SGInnovate in Singapore. The community event brought together perspectives on AI governance, agent security and the practical requirements of trustworthy deployment.
Privacy during the workflow
For Silence Laboratories, this is closely connected to the design of private agent collaboration. An agent may be authorised to carry out a task without needing access to every piece of information that helps determine the result.
That distinction matters when a workflow combines commercially sensitive context or data held by several organisations. A useful architecture should make the permitted computation clear and limit what each participant learns.
Connecting security and computation
Our work on AP3 addresses this layer through privacy-preserving computation. It is part of a wider set of requirements that also includes agent identity, permissions and operational oversight.
Participating in forums such as CyberSG Connect helps us keep those requirements connected to the questions being asked by Singapore’s security and AI communities. The goal is to make private collaboration a usable building block for enterprise teams.
