- kill chain
- The ordered path from first contact to demonstrated impact. We report chains, not findings — severity lives in the sequence.
- foothold
- The first reliable execution context inside scope — a shell, a token, a panel login. Everything after is escalation.
iam:PassRoleabuse- Assuming a more privileged cloud role via a service permitted to pass it. The single most common real pivot we demonstrate.
- canary record
- A pre-planted marker row or file used to prove access without touching real data. Proof without exfiltration.
- promptware
- Malicious capability delivered through prompts rather than executables — injection that plans, persists, and moves laterally.
- indirect injection
- Prompt instructions smuggled inside retrieved content — RAG docs, emails, tickets — instead of typed directly at the model.
- tool poisoning
- Malicious instructions hidden inside an MCP/agent tool description — the tool doc itself becomes the attack surface.
- confused deputy
- An agent or service tricked into using its ambient authority on an attacker's behalf. The deputy isn't compromised; it's compliant.
- rules of engagement
- The signed constraint set for a test — scoped assets, disabled classes, required human gates, and the kill switch terms.
- blast radius
- The demonstrated reach of a chain — what an attacker could actually touch, not what a score suggests.
- EASM
- External attack surface management — continuous enumeration of internet-facing assets. We treat its output as rumors until proven.
- false positive
- A reported weakness that never executes. Our bar: zero shipped, ever — a finding isn't a finding until it replays.
Speak offense.
The vocabulary we use in reports, on calls, and across this site — defined the way an operator means it.