AI Agent Threat Intelligence from ClawSecure

ClawSecure publishes AI agent threat intelligence on the malware families and campaigns that actually target AI agents, including ClawHavoc, the largest known malicious skill campaign targeting the OpenClaw ecosystem and a dataset ClawSecure produced from its own audits.

AI agent threat intelligence works differently from traditional threat intel, because the delivery mechanism is a component the user installs on purpose. There is no exploit to patch. A skill or MCP server arrives looking useful, survives a casual read, and then behaves differently once it holds permissions, or in a later version. That is why indicators for agent threats are behavioral as much as they are network-based.

The complete ClawHavoc indicator list documents the command-and-control callbacks, relay payloads, and memory-file harvesting that define the campaign. The ClawHavoc explainer covers how the campaign works end to end, for readers who want the narrative before the indicators. The sleeper agent problem covers the timing failure underneath all of it: a component that is clean when you scan it can turn malicious in a later version, which is what makes one-time verification insufficient.

ClawSecure owns this dataset rather than reselling one: the ClawHavoc indicators come out of ClawSecure's own audits of thousands of agent components, the same audit set in which 41% carried at least one security vulnerability.

ClawSecure's Watchtower re-checks tracked components after install and flags code changes, and its AI CISO blocks the malicious behavior these indicators describe while an agent is running, so a component that turns malicious after a clean scan is caught rather than trusted.