AI with the right to act: what is changing for information security?

At SOC Forum 2026, they discussed the security of AI agents, who not only answer questions, but also receive the right to act in corporate systems.

AI is increasingly performing tasks through connected tools: working with email and CRM, changing data, launching business processes. Access control remains necessary, but an additional task arises - to limit the actions of the agent, which selects them based on the request and incoming information. How to ensure that a technically permitted action complies with the user's task and business rules?

Date and time: October 28, 2026, 14:45–15:45 (Moscow time).
Location: Moscow, Lomonosov cluster, Ramensky Boulevard, 1; Hall "Atom".
Format: offline and online.

Discussion questions

1. AI agent: what rights can be delegated to it?

How does managing access for an AI agent differ from managing access for an employee or a regular application? Which agent protection tasks are already solved by IAM/PAM, and where additional control mechanisms are needed? At what point does an agent's autonomy become an unacceptable risk?

2. Who controls the AI ​​agent: the user or the data it reads?

Among the critical risks of AI agents are prompt attacks, data leaks when interacting with the Internet, and the possibility of fixing a prompt attack in the agent’s memory.

3. On what basis do we accept the model for corporate tasks?

By what criteria should a company decide that a particular model can be accepted for a particular corporate task?

4. Incident involving an AI agent: how to reconstruct events and stop the damage?

Are existing Incident Response processes ready to investigate incidents involving AI agents?

Recording and materials

The video has not yet been published. We will add it after it appears.

• Official program of SOC Forum 2026
• Announcement of participation in Telegram