Knowledge

Knowledge: the security of software agents in plain words

This section brings together analytical and legal materials on autonomous software agents: how an agent with tool access differs from a conversational model, how real-world incidents and attack vectors unfold, why text filters do not replace deterministic verification of tool calls, and which legal rules govern machine actions.

Topics

All pages by group

Where to start

How autonomous agents work, documented incidents of 2025–2026, enterprise deployment risks and agentic payment protocols.

How agents are attacked

Control hijacking through external data, vulnerabilities in the Model Context Protocol (MCP), risks in software pipelines and the threat database of the Federal Service for Technical and Export Control of Russia (FSTEC).

How agents are protected

Limits of manual approval and text guardrails, third-generation information security (IS 3.0), formal verification and cryptographic action receipts.

Rules and liability

Requirements of Federal Law No. 243-FZ, FSTEC Order No. 117 and personal data statutes, as well as corporate and executive civil liability.

Management and strategy

Cybernetic governance under Ashby’s law and Beer’s viable system model, scenarios to 2035 and verification of empirical claims.

Reference

Glossary of third-generation information security terms, the monograph “Proven Action” and the complete map of all sections.