The SOC analyst's job has been drowning in alerts for a decade — this is the first open-source attempt to build an AI agent that actually shows its work.
The Summary
- AiSOC is an open-source AI Security Operations Center that ingests security telemetry, correlates threats, and investigates them with AI agents whose reasoning is fully auditable
- Ships with 7,023 detection rules, runs locally with no API keys required, and records every agent decision in a replayable ledger
- Includes an MCP (Model Context Protocol) server for Claude, Cursor, and Continue — meaning the AI analyst can be steered by the same interface developers already use
The Signal
Security Operations Centers have an alert problem. The average SOC analyst sees 4,484 alerts per day according to IBM. They investigate maybe 10% of them. The rest pile up in dashboards until someone deletes them in bulk to make the red numbers go away. This isn't because analysts are lazy — it's because most alerts are noise, and the signal is buried under repetition, false positives, and vendor FUD.
AiSOC is trying to solve this with agents that triage, correlate, and investigate — but with one critical difference from every security vendor pitching "AI-powered threat detection." Every prompt the agent sends, every tool it calls, every reasoning step it takes gets recorded in what they call a "replayable decision ledger." You can literally watch the AI's work, step by step, and audit why it decided something was a threat or noise.
"New threat intelligence re-sweeps the history you already collected, and a human approves before anything reaches a vendor."
The architecture matters here:
- 7,023-rule detection library with 2,538 executable rules
- Pulls real threat intelligence (CISA's Known Exploited Vulnerabilities catalog) minutes after boot
- Maps investigations to MITRE ATT&CK framework
- Self-hostable, MIT licensed, runs without sending data to third parties
The MCP server integration is the sleeper feature. Model Context Protocol is Anthropic's open standard for connecting AI assistants to external tools and data sources. AiSOC shipping with MCP support means security teams can query their SOC's AI agent through Claude Desktop, or integrate it into their development workflow through Cursor and Continue. The agent that's triaging your alerts becomes queryable by your incident response team in the same interface they're already using to write code or investigate logs.
This is Web4 architecture in security operations. The agents run locally, they build context from your specific telemetry, and humans stay in the loop. But unlike traditional SIEM tools where the "AI" is a black box scoring alerts, every decision here has a paper trail. You can replay exactly what the agent saw, what rules it ran, and what reasoning led to the verdict.
The Implication
If you're running a security team, the question isn't whether AI agents will triage your alerts — it's whether you'll use auditable ones or vendor black boxes. AiSOC proves you can build agent-based security operations that show their work. The fact that it's MIT licensed and self-hostable means security teams can actually inspect what's happening under the hood, modify detection rules, and keep their telemetry inside their own infrastructure. Watch this space — the first wave of security AI was scoring and correlation. The second wave is agents. The teams that figure out auditability first will own the category.