On a monitoring quote, "24/7" can mean very different things. We judge it on 2 checkable points. The first is which log sources a person actually reads between 6 p.m. and 8 a.m. The second is which containment actions that person may take without phoning you. Public guidance from NIST and CISA speaks to both, and this post walks through each one.
The problem
Security tools keep producing alerts after the office closes, because endpoint detection, the firewall and Microsoft 365 all log around the clock. What usually breaks down overnight is the step after detection. NIST's SP 800-137 guidance, written in 2011 for federal systems, describes continuous monitoring as "maintaining ongoing awareness of information security, vulnerabilities, and threats to support organizational risk management decisions." A log that nobody opens until Monday gives you a record, not awareness.
The NIST Cybersecurity Framework 2.0 puts continuous monitoring under its Detect function, next to the analysis that decides whether an event is actually an attack. That second half is the part that needs people.
The 5 sources worth watching overnight
For a typical small office, these are the signals an overnight team should cover. If a provider can't confirm all 5, ask which ones are missing.
Microsoft 365 and Entra ID sign-ins, including new MFA devices and logins from places your staff have never been.
Laptops and servers running endpoint detection software, such as Microsoft Defender for Business.
Firewall and VPN logs, which show who is connecting from outside the building.
Email, where new forwarding rules and clicked phishing links tend to show up first.
Backup jobs, because attackers often go after backups before they encrypt anything.
The last one gets skipped more than it should. A backup job that suddenly fails, or a backup service that gets stopped, is often an early sign of an attack rather than a storage problem.
What a quiet night looks like in the logs
An overnight shift is mostly ruling things out, and it helps to know what normal noise looks like. A sales rep signs in from a hotel near DFW Airport after a late flight. A server restarts at 1 a.m. because a patch was scheduled. A backup job fails because a drive filled up. Each of those raises an alert, and each one needs a person to match it to something expected and close it with a note.
That triage only works when the analyst knows your baseline, such as who travels, which servers reboot on which night and which vendor connects remotely on Sundays. Without it, you either get paged for every hotel sign-in or the real alert sits in the queue.
How a real alert plays out
Here's the pattern for one common attack, written as an illustration rather than a real incident.
Shortly before 2 a.m. on a holiday Saturday, the endpoint software on a file server reports a new local administrator account created during a remote session nobody scheduled.
A few minutes later, the same session tries to stop the backup service and delete the server's restore points.
The analyst confirms no maintenance was planned, isolates the server from the network with the endpoint tool and disables the new account.
The analyst calls the first named contact to explain what was contained and what still needs a decision.
On Monday there's a written timeline, including next steps like resetting the remote access credentials that were used.
Those first 2 signals are the useful pattern to remember. A new local admin account plus an attempt to remove restore points is a combination that ransomware operators often produce before encryption.
The containment actions to pre-approve
CISA's #StopRansomware Guide, updated in 2023 with the FBI, NSA and MS-ISAC, starts its response checklist with identifying the affected systems and isolating them right away. That only works overnight if the analyst already has permission. These are the actions a typical agreement pre-approves.
Isolating a laptop or server from the network.
Disabling a user account.
Blocking an address at the firewall.
Then write down the exceptions. A payroll server or a lab instrument that has to keep running should be listed as hands-off unless someone senior agrees.
Gotchas
An alert-only plan isn't monitoring. It forwards the alert and stops there, so the containment step never happens at night.
A call list with one name fails on holiday weekends. Use 2 named contacts with 2 phone numbers each.
"As soon as possible" isn't a response time. Get a number of minutes from alert to first action for high-severity alerts, in writing.
A daily raw export of alerts means your own staff are doing the triage at 8 a.m.
Holiday weeks are the stress test. If an office closes Thursday and Friday of Thanksgiving week, 138 of the week's 168 hours go by with nobody at a desk.

Backend
Source:https://dev.to/upriteservices/what-247-monitoring-should-watch-and-what-it-should-be-allowed-to-do-51nn
What 24/7 Monitoring Should Watch, and What It Should Be Allowed to Do
dev.to Security0 views

PreviousTUN mode says connected but nothing loads — debugging five layers in orderNextKeep license tools away from your AI agents, and reconcile MSP seats in plain code instead
Related posts
The Four Product Risks: What Counts as Evidence for Each
0dev.to UX
Ghost jobs in 2026: how many job postings are real?
0dev.to Data
Merging and deduplicating results from several scraper runs (without a new script each time)
0dev.to Data
WARN Act and Layoffs in Wholesale Trade — What Distribution Workers Should Know
0dev.to News
Dev News Digest: 16 Nov 2026, 18:00
0dev.to News
Developer or designer for an AI-built app: How to tell which problem you have
0dev.to Design