“Where does a wise man hide a leaf? In the forest.” G.K. Chesterton’s line is still the best short description of deception security. Instead of only building walls, you fill the network with fake servers, fake credentials and fake files. Real users never touch them. Anyone who does is almost certainly an attacker, which makes a decoy alert about the cleanest signal in security.
The idea is old. In 1944 Operation Fortitude used dummy airfields and inflatable tanks to mislead German commanders about the Normandy landings. Decoy networks apply the same trick to IT. What has changed by 2026 is who buys it, who sells it, and a new target: AI agents.

How Decoy Networks Work
- Decoys. Fake servers, databases, OT controllers or cloud workloads that look real and sit next to production systems.
- Lures and breadcrumbs. Fake credentials, browser sessions and mapped drives planted on real endpoints that point attackers toward decoys.
- Honeytokens. Fake API keys, documents or database rows that raise an alert the moment anyone uses them.
- Engagement. Some platforms let the attacker keep working inside the decoy so defenders can study tools and intent.
The doctrine assumes some breach is inevitable. The goal is to catch attackers early, during lateral movement, and to make every step more expensive for them.
The Market Consolidated
The pioneers mostly ended up inside bigger platforms. SentinelOne bought Attivo Networks in 2022 for about $617 million, mainly for its Active Directory and identity protection. Zscaler bought Smokescreen, Proofpoint bought Illusive and Commvault bought TrapX. Deception is now often a feature of endpoint, identity or zero trust suites rather than a separate purchase.
| Vendor | Status in 2026 |
|---|---|
| Attivo | Part of SentinelOne (identity and deception) |
| Smokescreen | Part of Zscaler |
| Illusive | Part of Proofpoint |
| TrapX | Part of Commvault |
| Acalvio | Independent, enterprise and AI agent deception |
| Thinkst Canary | Independent, low-cost canaries and tokens |
| CounterCraft | Independent, threat intelligence focus |
The New Front: Deception for AI Agents
AI agents read files, call APIs and use credentials on their own. That makes them easy to fool, and dangerous when compromised. In July 2026 Acalvio launched what it calls Deception Guardrails: honeytokens planted in files and configs that agents read, plus decoy MCP servers, decoy retrieval systems and decoy agents around a company’s real AI stack. If a hijacked agent reaches for a fake tool or credential, defenders know at once. In September it added an agent that writes deception playbooks from plain-language requests inside Google’s Gemini Enterprise marketplace.
The same logic works against attackers’ own AI tools. Automated reconnaissance trusts what it finds, so a network full of plausible decoys wastes the attacker’s compute and time.
Where Deception Fits Best
- Identity attacks on Active Directory and cloud identity, where fake accounts catch credential theft.
- OT and critical infrastructure, where you cannot install agents on old controllers but can place decoy PLCs on the network.
- Ransomware early warning, by catching the scanning and lateral movement before encryption starts.
- AI infrastructure, the newest use, covered above.
More security coverage is in our Cybersecurity and Defense section and at cybersecuritymarket.com. For the AI side of the threat, see our piece on AI model prices and data security funding.