There's a difference between a tool that bolts an LLM onto an existing workflow and one architected around AI-assisted operation from the first line of code. Erebus is the second kind, and that's not a marketing distinction. It changes what the system can and can't do.
What AI-native actually means here
In Erebus, the AI reads teamserver state, reasons over active sessions, and prepares the next task as the primary control loop, not as a side panel summarizing logs. Everything downstream, from the task queue to the approval gate, was designed around that loop existing from day one.
Human-in-the-loop by design, not by disclaimer
Approval gates, not warnings
Sensitive actions don't just log a warning. They stop and wait for the operator to approve, deny, or redirect. The AI can plan; it can't execute high-risk steps alone.
Secure infrastructure as a baseline
Authenticated operator access and encrypted implant communications aren't optional add-ons. They're core requirements the framework was built against.
Open source, so you can verify it
Don't take "human-in-the-loop" on faith. The code is public. Read the task queue, read the approval logic, confirm the AI can't route around the gate.
Why this matters for AI security tooling generally
New AI-powered security tools face a trust problem: they're new, they're often closed-source, and they connect to external services in ways that are hard to audit. Erebus makes the opposite bet: open source, modular, cross-platform, and built so the human stays the approval authority on anything that matters.
The point was never unattended exploitation. It's supervised speed: an AI that moves the operation forward and a human who still understands, and controls, every high-risk step along the way.