Guide, read 10 Nov 2026
Authority to act is the list of actions the provider may take without calling you, and you set it in writing
What does authority to act mean in an MDR service?
Authority to act is permission, given in advance, for named actions on named systems. Without it, an analyst who finds ransomware at 3 a.m. phones your on-call engineer and waits. With it, the analyst isolates the host and then phones. The phrase covers three parts: the action, the target and the condition.
Providers publish their action lists. Critical Start lists endpoint actions as isolate host, block file hash and forensic capture, and identity actions as reset password, disable account and re-initiate MFA. Deepwatch lists session revocation, password reset and account control for identity, and process termination and host isolation for endpoints. Microsoft lists isolate device, quarantine file, restrict app execution, release from isolation, remove app restriction and disable user. These are the providers' own lists, read on 10 Nov 2026.
The three modes providers describe
Published pages describe three modes. In automatic mode the provider acts and tells you. In approval mode it asks first. In monitor-only mode it alerts and does not act. Deepwatch says its Active Response is opt-in and that actions run automatically, with analyst approval or in monitor-only mode, as set by the customer. Red Canary says customers can keep human oversight on specific response actions or choose to automate. Expel says customers can let it press the respond button or choose what, how and when it acts.
Microsoft ties the mode to a permission setting. Its page says experts act on the customer's behalf only if granted Security Operator access; with the default Security Reader access, actions are left as pending for the customer. A service can therefore be bought as managed response and run, by default, as advice.
How to set it
Build a small matrix. Rows are asset classes: laptops, servers, domain controllers, executive accounts, production systems. Columns are actions: isolate host, kill process, disable account, block address, revoke sessions. Each cell holds one word: automatic, ask first or never.
Start where the damage from waiting is high and the damage from acting is low. Isolating a laptop is reversible in minutes. Disabling a service account can stop an invoice run. Isolating a database server can stop a shop. eSentire describes its response as policy-bounded and reversible, which is the standard to ask of any provider: every automatic action needs an undo you can name.
- Laptops: isolate and disable the user automatically; notify at once.
- Servers: isolate automatically only outside business hours; otherwise ask first.
- Domain controllers and identity systems: ask first, with a named approver and a back-up.
- Production and OT systems: never automatic; advice and escalation only.
- Executive accounts: disable the session, ask before password reset.
Writing it into the contract
Put the matrix in an appendix and refer to it from the service description. Name who may change it, by what route and how fast the provider must apply a change. Ask for a log of every action taken: time, target, analyst, reason. Critical Start says its analysts run the customer-approved playbook and that notice ahead of action is available if the customer prefers, which is an example of authority written as a playbook and not as a promise.
Include a stop word. If your team calls a halt during an incident, the provider must stop automatic actions at once.
Testing and review
Test within the first two weeks. Create a harmless alert on a test laptop and watch what happens: who is called, which action is taken, how long it takes, and whether the undo works. Repeat the test whenever you change tools or staff. Review the matrix every quarter, and after any incident, against what the provider actually did. If the provider acted outside the matrix or waited when it should have acted, fix the paper first and the people second.
Account actions deserve particular care because stolen credentials are a common route in. MITRE ATT&CK describes adversaries obtaining and abusing credentials of existing accounts for initial access, persistence, privilege escalation or defense evasion (technique T1078). Disabling an account is often the fastest way to stop that.
Related pages
- All guides
- MDR vs MSSP, the full comparison
- MDR glossary
- The MDR Service Matrix
- Kinds of managed detection service
- The first 30 days of an MDR service
- What to check in an MDR contract
Sources
- MITRE ATT&CK, T1078 Valid Accounts, read 10 Nov 2026
- Critical Start, MDR page, read 10 Nov 2026
- Deepwatch, Active Response, read 10 Nov 2026
- Microsoft, Defender Experts managed response, read 10 Nov 2026
- Red Canary (Zscaler), MDR page, read 10 Nov 2026
- Expel, auto remediation, read 10 Nov 2026
- eSentire, response and remediation, read 10 Nov 2026
Last reviewed 10 Nov 2026.