Skip to content

ThreatOps FAQ: Straight Answers for Infrastructure Teams

Published November 15, 2026 by Beena Yadav

People who run infrastructure keep asking the same handful of questions about ThreatOps, usually in the middle of a planning meeting. Below are straight answers, drawn from the Ops Atlas entry on the discipline, with no marketing gloss. Where the Atlas is uncertain, I say so.

What does ThreatOps actually cover?

The Atlas frames it as gathering intelligence about adversaries and searching your own estate for them, handled as a single joined-up job. That means written requirements, rated sources, analysis and reporting, handling of indicators and attacker behaviours, standard exchange formats, intelligence platforms, hunting itself and community sharing. It does not cover production alert rules, round-the-clock alert triage, incident handling, vulnerability prioritisation or telemetry plumbing; the Atlas assigns those to neighbouring disciplines. One handoff ties the neighbours together: a hunt result that holds up gets passed on as a request for a lasting alert rule.

Is it an official standard?

No. According to the entry, there is no formal definition from any standards organisation, and the NIST glossary lists nothing under that name. It labels the term ambiguous and notes that the practices inside it are well established. Vendors also use the same word for products and team names, so check what a supplier means before you sign anything. Official guidance exists around the pieces: a 2016 NIST publication on information sharing, a 2024 revision of the cybersecurity framework with controls for risk assessment and adverse event analysis, and a hunting control in the NIST catalogue that the Atlas points out is optional and appears in no baseline.

Is a threat feed the same thing?

No, and this is the most expensive confusion. A feed is an input. The Atlas quotes the incident-response community’s view that feeds exist to raise flags and alerts, and that they are unsuited to automatic blocking. ThreatOps is the surrounding work: deciding what you need, rating the feed, expiring entries and measuring whether it was ever useful.

How many questions should we track?

Five to eight at a time. Every requirement is limited to a single question, the one fact that would answer it and the one decision it supports, and the whole set gets a quarterly review. More than that and nobody can say which ones the team is working on. Broad topics must be split up: “everything about ransomware” is a theme, not a requirement.

When can we let a feed block traffic automatically?

Only after measuring. The Atlas sketches a sample rule with four conditions: the provider scores A or B for reliability, the item scores 1 or 2 for credibility, ninety days of results show the feed at 0.9 precision or above, and nothing on your allowlists matches. Whoever is on call can revert any block, and the owner reviews it within an hour. If a feed fails the precision test, it keeps alerting only. A shared content delivery address that gets blocked by mistake is the cautionary example the Atlas uses, and its runbook calls for reverting in minutes.

How long should an indicator live?

As long as you decide. The standard exchange format makes the expiry field optional, which means missing expiry equals no limit. The Atlas’s sample policy gives addresses fourteen days, URLs at thirty, domains at sixty and file hashes at a year. Pick your own numbers, write them down and automate the cleanup.

How often should we hunt?

About once a fortnight, and the Atlas keeps each session to a pair of behaviours at most. It recommends a hypothesis structured around four prompts: actor, behaviour, location and evidence. Every hunt ends with an artefact, and the pass mark is a colleague who sat out the hunt repeating it from your notes and landing on the same answer.

What if a hunt finds nothing?

Then check your data. If the records were good enough that the activity would have appeared, you have a confident negative, which the Atlas counts as a successful hunt. If the records were incomplete, you write a gap ticket and fix the logging. Reporting “not affected” after merely sweeping for known indicators is the mistake it warns about, citing a 2024 joint government advisory on a state-linked intrusion set where conventional indicators were described as generally lacking and footholds were held for at least five years.

What does a good hunt record look like?

Short and boring. Name the hypothesis in two sentences, list the data sources you searched and how far back they reach, say what turned up, and close with one of the four endings. Add the date of the next review. If a new hire can follow the page without asking you questions, the record passes. If you need to explain it over a call, it needs more detail, because the entire point of an artefact is that it outlives the person who wrote it.

Is attribution worth the effort?

Rarely, for operations teams. The Atlas calls out attribution theatre as an anti-pattern and explains that group data from the public attacker catalogue is background rather than a forecast. In its lab comparison, three well-known groups drew on ninety-two different techniques in total, and just two techniques showed up in every one of them. Knowing the name of a group does not tell you what it will try next, and the hours spent on naming are usually better spent on checking your own logs.

Do the numbers support doing this?

The Atlas leans on the 2026 edition of a large annual incident-response report. It shows median dwell time worldwide creeping up from eleven days to fourteen, with espionage-type and certain insider-style cases stretching to 122. Voice phishing ranked second among initial entry routes, at eleven percent. These are context for why looking inside your own environment matters, not promises about any particular organisation.

Can a small team do this without buying a platform?

Yes. The Atlas suggests a version-controlled repository as a starting point, and the tooling choice comes after three gating questions. Missing ninety days of logs? Repair that. No owner for findings? Hire hunting as a service. No stakeholders asking for intelligence? Keep an advisory setup. Pass all three and you build. Open-source platforms such as MISP and OpenCTI are there when manual handling starts to drag.

What is safe to share, and with whom?

Use the colour labels of the traffic-light protocol and agree categories in advance. Sharing communities, often organised by sector, are where most small teams exchange warnings. The Atlas also flags a date to check: legal protections for sharing in a US statute from 2015 expire on 11 December 2026, so confirm the current position with your own counsel. This post is general information, not legal advice. In Europe, the NIS2 directive encourages voluntary sharing in its Article 29.

How do we measure success?

Four ratios do the job: how much of your requirement list is covered, how much intelligence becomes action, how many hunts convert, and how many gaps close. In the Atlas’s sample quarter they land at 83, 35, 42 and 44 percent. Skip the vanity counts of indicators ingested, reports produced and hunts started.

Does this help my own career?

If you come from a sysadmin background, the habits transfer well: reading logs, spotting baselines, documenting. Our notes on moving from sysadmin and support into DevOps describe the same shift in mindset, and a roadmap of skills, projects and certifications for engineers is a good place to slot in a small threat-hunting project.

Where is the source for all this?

Every answer above comes from the ThreatOps entry in the Ops Atlas. It runs long, so skim the headings and treat its maturity ladder and its 30-60-90 schedule as your checklist. Pick the question your team argues about most, and see whether the entry settles it.