Fresh advice for a fast-moving area
Kaspersky has released practical guidelines on keeping autonomous AI agents safe inside company networks. The document draws on real incidents and threat models to show how organisations can limit damage when these systems go wrong or get compromised.
The guidelines were presented at the AI Everything Global event in Abu Dhabi on 6-7 October. They focus on the Cyber Immunity approach, which builds security in from the start rather than bolting it on later.
Core design rules
The report lists four main principles:
- Treat large language models and any external data as untrusted by default, and require human approval for irreversible actions.
- Keep the trusted computing base as small as possible.
- Isolate components using sandboxes and virtual machines.
- Apply a default-deny policy to tool calls, network traffic and supply-chain changes.
It also urges CISOs and CTOs to maintain an inventory of AI agents, set clear isolation policies and manage the agent supply chain.
Why it matters now
AI agents already handle tasks from routine automation to software development and security operations. As these tools gain more autonomy, the risks of excessive privileges or data loss grow. Kaspersky’s head of AI Technology Research, Vladislav Tushkanov, said: “While Cyber Immunity was initially conceived outside the realm of AI, its core principles map directly onto advanced LLM‑based systems. When developers work with fundamentally non‑deterministic and untrusted components — which large language models should be considered by default — they should embed trust into the solution’s architecture to limit the potential impact of errors or compromise.”
We covered Kaspersky’s recent AI updates in threat intelligence just yesterday. For broader VPN and privacy tools that can help protect everyday browsing, see our VPN section.
