Answers grounded in
your security findings.
Ask questions in plain language and review answers alongside the scan findings that support them. Optional, off by default, and powered by the AI provider and key you choose.
- Write permissions
- 0
- Your choice of provider
- Your key
- Live read-only tools
- 12
The problem
A list of findings is not the same as knowing what to do.
Turn a finding into a useful question: why does it matter, what should we review, and where can we start? The assistant helps explain your results while keeping the supporting evidence available for your team to check.
What you can ask
Plain language. No query syntax to learn.
- “Which of these findings should I fix first?”
- “Which MFA methods has Dana registered?”
- “Why does this one matter — what would an attacker do with it?”
- “Walk me through fixing the SharePoint sharing issue.”
- “What changed in my tenant this week?”
- “Which apps have access to my tenant, and what were they granted?”
How it works
Your scan results, with supporting sources.
- 01
It reads your scan
Answers come from your own completed scan — the findings, their severity, the evidence behind them. Open the linked findings to check its interpretation against the saved evidence.
- 02
Optionally, your live tenant
Turn on live access and it can also read current state through its own read-only app: which MFA methods are registered, which supported settings changed, and what the available evidence shows. Registration does not prove enforcement. Off by default.
- 03
It checks Microsoft's docs
Before giving you portal steps, it looks up Microsoft's current documentation. Menus and blades get renamed constantly, and a walkthrough from stale training data sends you somewhere that no longer exists.
Where your data goes
Understand what your AI provider receives.
Review how identifiers are handled, which data is shared, and the controls available before you enable the assistant.
Your provider, your key
You bring your own OpenAI, Azure OpenAI, or Anthropic account. Your findings go to the model you chose, under your agreement with them. Tenant Strike never sends your data to an AI provider on its own behalf, and we never see your key after you save it.
Names replaced before sending
Email addresses, IP addresses and display names are swapped for placeholders before anything leaves our servers, and swapped back on your screen. The model reasons about the finding; it does not learn who your people are.
Read-only, and separately revocable
Every permission it holds ends in .Read. It cannot change a setting, a user, or a policy. Live tenant access runs through its own app registration you consent to separately — so its reads appear under their own identity in your sign-in logs, and revoking it does not stop your scans.
Review available request diagnostics and the published API inventory to understand the assistant’s access separately from the scanner’s. See the read-only assurance page.
What it won’t do
Keep the evidence and its limits in view.
Missing permissions, unavailable data or a throttled request can limit an answer. The assistant is designed to explain those gaps and link to supporting findings. Review its interpretation before acting; an AI answer is not proof that a control is effective.
Its connected tools are read-only. Your team makes the changes.
AI Assistant · Included with Pro · Beta
Ask a useful question. Start with your findings.
Included in Pro at no extra cost. Bring your own AI provider key — or leave the assistant off entirely and the scanner works exactly as before.