Detection · Incident response
EDR, XDR, SOC: who responds at 3:12 am?
Published7 min read
An alert can be accurate and still go unanswered. Before comparing cybersecurity offers, I would check who receives it, what they can decide and who takes over at night.
The article in 30 seconds
- EDR and XDR are tools. A SOC is a team with procedures. MDR is a detection and response service.
- Detection is only the first step. Someone needs to assess the alert and be able to act, including at night and on weekends.
- The contract and the trial make the difference. Check what is monitored, who can intervene and how long a response takes in practice.

Saturday, 3:12 am. The tool installed on the servers reports that an administrator account is running an unusual script against the file share.
The alert is accurate. Who will read it, and who can isolate the server before encryption spreads?
This fictional scene raises a question the acronyms sometimes obscure: who acts on an alert, with what authority and at what time?
Tools and the teams that operate them
Vendors list EDR, XDR and SOC as if they were three tiers of a product. They address different needs.
| Acronym | What it does | Where it stops |
|---|---|---|
| EDR | Monitor workstations and servers that have an agent, detect suspicious behaviour and apply responses. | Devices without an agent and events outside the configured rules. |
| XDR | Connect several sources, such as endpoints, email, identity, cloud or network data. | Sources that are not connected and business decisions outside its remit. |
| SIEM | Centralise and correlate security logs. | Log quality, maintenance of detection rules and the integrations needed to act. |
| SOC | A team and procedures for monitoring, analysis and coordination of response. | What its tools can see, the hours it covers and the permissions it has. |
| MDR | A provider operates detection tools and responds to alerts. | The scope, hours and actions set out in the contract. |
What do these acronyms mean?
EDR: Endpoint Detection and Response, for detecting and responding to threats on endpoints. XDR: Extended Detection and Response, which extends detection and response across sources. SIEM: Security Information and Event Management, for managing security information and events. SOC: Security Operations Center, the team and its operations. MDR: Managed Detection and Response, a service that runs detection and response for you.
EDR and a SOC play different roles. EDR can respond automatically. A SOC handles ambiguous cases and coordinates escalation when business operations are at stake.
In Microsoft Defender for Endpoint, for example, automation is configured by device group, with options ranging from automatic remediation to approval before acting. [1] People still decide what the tool may do on its own.
From alert to action: three steps to connect
An automatic response can start as soon as a threat is detected, if it has been authorised and configured. The diagram describes who is responsible. Individual actions need not all wait for a person.
XDR can add context. An email login using a stolen password may look very different from malware on a workstation.
Of the 661 cases in Sophos’s 2026 report, about 67% of intrusion causes were identity-related. [2] The service still needs to connect the relevant email systems and accounts. The included connectors tell you more than the word “XDR” on a brochure.
At Target, the alert had already been raised
In late November 2013, Target’s FireEye system detected malware used to exfiltrate data. A US Senate report noted unanswered alerts and automatic removal that had not been enabled. About 40 million payment card accounts were reported compromised. [3]
In 2017, the retailer reached an $18.5 million settlement with 47 states and the District of Columbia to close their investigation. That figure does not represent the total cost of the incident. [4]
The lesson goes beyond product choice: buying detection also means organising triage, decisions and the authority to act.
The real test starts after closing time
In the cases Sophos studied, 88% of ransomware deployments and 79% of data exfiltrations occurred outside business hours. The median time before an attempt to access Active Directory fell to 3.4 hours. [2]
These figures come from cases handled between November 2024 and October 2025 by a company that sells these services. They do not describe every organisation. They do raise a practical question: who is on duty while your team sleeps?
Why round-the-clock coverage costs more than a subscription
Keeping one seat staffed continuously takes 8,760 hours in a 365-day year. At the French benchmark of 1,607 hours of actual work per year, that is about 5.5 full-time equivalents. [5]
This is a rough figure. Staffing also has to cover additional absences, training, supervision and specialist expertise. Sharing coverage through a provider can therefore make sense.
I would start with the team you already have
Company size alone is insufficient. I would first look at the people available, the critical systems and the hours to cover. Here are starting points to adjust to your risk assessment.
| Your situation | What I would consider | What to check |
|---|---|---|
| No security team | EDR operated by an MDR provider available around the clock. | Your managed IT provider should not be the only recipient of an overnight alert. |
| Team available during business hours | EDR or XDR, with MDR coverage at night and on weekends. | Who makes decisions during the handover? |
| Email and cloud services are critical | XDR connected to identity and email systems, operated by an internal team or MDR provider. | Which sources the licence actually includes. |
| Varied sources and a need to keep logs | EDR and, if justified, SIEM operated by a SOC. | The budget for maintaining collection and detection rules. |
Without a security team, I would start by checking multifactor authentication, protected and tested backups, and whether anyone can actually monitor the EDR before buying XDR or SIEM.
Four questions to put in the contract
1. Who can disconnect what?
Isolating a salesperson’s laptop and taking a till server offline on a Saturday have different consequences. The contract should identify authorised actions, those requiring approval, the devices covered and the backup contact. It should cover automated actions and how to reverse them after an error.
2. Which delay is measured?
Receiving a notification does not mean the incident is under control. I would ask how long it takes to assess and contain an incident, with anonymised examples from both day and night.
3. What happens if the tool goes quiet?
Which workstations, servers, hypervisors and cloud accounts are outside scope? Who watches for loss of telemetry? ESET studied nearly 90 tools designed to disable protections. [6] An agent suddenly going silent should also trigger an investigation.
4. Who keeps the logs?
Where are logs kept, for how long and who can access them? The company should know how to retrieve them in a usable format, both during an investigation and when the contract ends.
Staff monitoring and qualifications: check the scope
EDR can collect records of activity on workstations. Review its configuration and use with the data protection officer or a lawyer: define the purpose, keep collection proportionate, set limits on retention and access, and inform staff beforehand. For systems that monitor staff activity, the CNIL also points to consultation with the social and economic committee in private companies with 50 or more employees. [7]
For a sensitive system or a specific requirement, I would verify the service’s qualifications. ANSSI distinguishes between qualified detection providers (PDIS) and qualified incident response providers (PRIS). Check which specific service and activities the qualification covers. [8]
NIS 2 · Status as of 8 November 2026
France’s National Assembly announced that it would begin examining the resilience and cybersecurity bill on 9 November. The timetable should be checked again after that date. [9]
If an offer is described as “mandatory for NIS 2”, I would ask for the legal text and the specific measure the offer covers. A product acronym alone cannot establish organisational compliance.
Before signing, run the Saturday night test
I would run a trial in a real but limited environment, with written authorisation, an identified test owner and stop conditions. Agree on the expected response to each scenario. A script that merely looks unusual is not necessarily an attack.
| Authorised scenario | What to check |
|---|---|
| A harmless script reproduces an agreed technique | The advertised detection, the analysis and the incident report. |
| A test account signs in from an unusual location | The expected identity response, if the source and rule are covered, followed by verification with the company. |
| The agent stops on a test device | Detection of the missing agent and contact within the agreed time. |
I would record each timestamp: when the alert appeared, when the call came, who answered, when the decision was made and when isolation took place, if it did. Monday’s report should explain those steps in language a manager understands.
I would compare providers on these responses, for the same scope, before comparing prices.
Back to Saturday at 3:12 am. If the company knows who will see the alert, what the tool can do on its own and who decides when operations are at stake, the choice of acronym becomes a technical matter.
The right service is one whose response you can describe, even in the middle of the night.
Sources and references
- Microsoft Learn — Defender for Endpoint remediation and automation levels ↩
- Sophos — Active Adversary Report 2026: findings and methodology ↩
- US Senate — analysis of the 2013 Target breach (PDF, 2014) ↩
- Tennessee Attorney General — $18.5 million settlement with Target (2017) ↩
- Insee — statutory working time and the 1,607-hour annual benchmark ↩
- ESET — research on tools that disable EDR (19 March 2026) ↩
- CNIL — monitoring employee activity (9 July 2026) ↩
- ANSSI — PDIS and PRIS qualification standards ↩
- French National Assembly — 4 November 2026 sitting and agenda announcement ↩