Skip to content
Tech Interview Prep
Technical interview guide

Security Monitoring & Logging

What to log, how SIEMs correlate it, and the difference between detection and prevention controls.

Read
30 min
Practice MCQs
25
Interview Q&A
25
Edition
v3
Editorial status
Reviewed

Scope: NIST SP 800-53 Rev. 5; NIST SP 800-92 final plus Rev. 1 initial public draft context; MITRE ATT&CK detection strategies accessed 2026-08-31.

Curated: · Written: · Reviewed:

25 questions

MCQ-23A team's CloudTrail trail has management events enabled for all regions, data events disabled, and log file integrity validation turned on; no other logging is configured. Reconstructing an incident that used a stolen `svc-deploy` key, they find four actions in the timeline: `AssumeRole` into `prod-admin` at 03:12, `GetObject` on `s3://payroll-exports/2024-q4.csv` at 03:14, `PutBucketPolicy` on the `payroll-exports` bucket at 03:16, and `CreateAccessKey` for a second IAM user at 03:18. Which of the four actions will be missing from the trail's logs?
MCQ-22A contractor's phished credentials were used to sign in to the internal reporting portal, and the intrusion went unnoticed for nine days. The team logs failed authentication attempts only; the application's own audit log records which records were read but not how the session was established. Which logging change lets an investigator reconstruct that session?
MCQ-21An EDR alert flags a signed binary on a Windows 10 22H2 host for suspicious activity against lsass.exe. The host runs Sysmon v15 with process access logging enabled and Windows Security object-access auditing configured. Which telemetry most directly establishes whether the binary read LSASS memory rather than only querying the process?
MCQ-25An analyst runs a 30-day query for a known lateral-movement binary across EDR process-creation telemetry that 92% of endpoints forward, and gets zero rows. The CISO asks what the incident report can state. Which conclusion is defensible?
MCQ-6A collector stamps every log line with `receipt_time` from its own clock and preserves the source's `event_time`. During an incident review you pull a 10-minute window of events from three hosts. Host A: 1,200 events, `receipt_time − event_time` = 3h 58m 12s ± 2s. Hosts B and C: 1,400 events each, the same difference = 1.4s ± 0.3s. All three run the same agent build. What does host A's data support?
MCQ-7A SOC consolidates 22 sources into one store. A detection joins IdP sign-in events to EDR process events on the `user` field, within a 5-minute window. The IdP emits `alice@corp.example`; the EDR emits `CORP\alice`. Both sources are ingested, and host clocks are NTP-synced to UTC. Over a 5,000-sign-in pilot, the rule matched 0 events. Which change is most likely to make it match?
MCQ-8A parser change renamed `user_id` to `userId` in indexed documents. Ingest counts matched the previous week, freshness stayed under a minute, and the search API returned 200, but correlation rules keyed on `user_id` stopped firing and analysts found the gap four days later. The team can add one check. Which check catches this class of failure?
MCQ-10A SIEM parses 40 log sources into versioned normalized fields and deletes each raw event 24 hours after it parses successfully. A defect in the ASA parser (v2.4) mislabeled `deny` as `allow` from 09:00 Monday until it was caught at 09:00 Wednesday. Investigators need to know which connections were truly denied across that 48-hour window. What can they recover?
MCQ-3Checkout requests enter through a gateway that mints a `request_id`, then fan out to an orders service and a payments service. Orders mints its own `order_uuid` after validation; payments writes a free-text line with a local timestamp. The gateway and orders run in us-east-1, payments in eu-west-1, and the three teams share no database. During a 40-minute incident 3,100 checkouts failed, and the team could tie only 18% of them back to the gateway request that caused them; 900 of the failures died in orders validation before an order was ever created. Which schema change makes all 3,100 failures reconstructable end to end?
MCQ-2A payments API writes one structured record per request (key ID, source IP, response bytes) to a central store with 90-day retention and no sampling. An attacker used a stolen API key for 9 days and moved 40 GB out; a customer reported the breach. The review confirms every exfiltration request is in the store, complete and searchable. Which gap does this incident demonstrate?
MCQ-4A service writes one JSON line per request. Before writing, a redaction pass walks the object recursively and replaces the value of any key whose name matches, case-insensitively, one of `password`, `token`, `secret`, `authorization`, or `dsn`: the whole key name, not a substring. String values are never parsed. Which emitted line still carries a live credential in plaintext?
MCQ-20An application writes one line per request to a log file using the template `user=<User-Agent> action=login status=ok`. A shipper splits the file on newline and indexes each line, and a SIEM rule alerts on any line containing `user=root action=login status=ok`. An attacker sends a request with the User-Agent `Mozilla/5.0\nuser=root action=login status=ok`, and the SIEM fires a false alert. Which change stops the forged alert?
MCQ-5An audit service writes one row per event to a Postgres 15 table `audit(id bigserial primary key, ts timestamptz, actor text, action text, payload jsonb)`. The service account has INSERT only. Two DBAs connect with superuser rights, which they need for migrations. Last quarter a DBA ran `UPDATE audit SET payload = ... WHERE id IN (...)` on 3 rows to hide a bulk export; it was found 11 days later in a manual review. The team must detect any change to an existing row's contents within 1 hour, and the DBAs keep their current rights. Which change meets that requirement?
MCQ-19Checkout spans carry the customer e-mail (the only stable customer identifier in the flow), the first six and last four card digits, and the raw request body, 8 KB on average. The service emits 40 million spans a day into a warehouse where 300 analysts hold SELECT on the span table, and spans are kept 30 days. Fraud must count failed checkouts per customer over a rolling 90 days, and privacy requires that an analyst querying spans cannot work out a customer's e-mail from them. Which change meets both?
MCQ-18A payments platform ingests 1.5 TB/day of security logs. Payment and authentication sources (0.5 TB/day) are PCI DSS in scope; the other 1.0 TB/day is internal telemetry with no external retention obligation and a standing instruction to minimize its retention. Hot storage costs $0.03/GB-month, cold archive $0.004/GB-month. PCI requires 12 months of retention with the most recent 3 months immediately available. Detection queries only the last 30 days; 3 of the last 4 investigations needed 6–9 months of telemetry. Which retention policy fits?
MCQ-1A payments platform's SOC runs a SIEM at its 1 TB/day ingest cap. A post-incident review of a 9-day account-takeover intrusion found the decisive evidence, the attacker's `AssumeRole` calls and the later use of a Kubernetes service-account token, would have been in CloudTrail and the Kubernetes audit log, neither of which is connected. The team measures that CloudTrail data events would add 400 GB/day and Kubernetes audit at RequestResponse level 300 GB/day, but it has only 250 GB/day of headroom. What should the team do first?
MCQ-24A cloud audit pipeline ingests 12 TB/day of CloudTrail events (40,000 events/s at peak). To cut cost, the team samples every event class uniformly at 1:50, bringing ingest to 240 GB/day. A privilege-escalation rule matches CreateAccessKey followed by AttachUserPolicy by the same principal within 10 minutes; it used to fire on about 30 real chains a day and now fires roughly once a quarter. The rule reads 40 of the 1,200 event classes in the schema, and those 40 are about 0.5% of ingest volume. Which change restores the rule's recall while keeping ingest under 500 GB/day?
MCQ-9A bastion host's auth log forwards to the SIEM. Over 90 days the source averaged 38 events/hour from 08:00–20:00 and 3 events/hour overnight, with the longest overnight gap between consecutive events at 2 hours 30 minutes. The current rule pages whenever the source sends zero events in a 10-minute window, so it fires most nights and analysts auto-close it. Last Tuesday the appliance's log partition filled at 02:10 and its writer stopped; the forwarder stayed connected with an empty queue and nothing arrived until 08:40. The gap was found three days later in a manual review. Which change best fixes this?
MCQ-13A detection rule fires when 5 or more failed logins for one account occur within a 60-second window. Backtesting over 30 days of auth logs, it fires on 38 of 40 labeled brute-force campaigns and on 240 of 12,000 benign login bursts. Replaying the same 30 days through the production pipeline, where auth events reach the index with a p50 delay of 20 s and a p99 of 150 s, it fires on 19 of the 40 campaigns and 90 benign bursts. Which conclusion does the replay support?
MCQ-11A team keeps its detection rules in Git and deploys them through a pipeline. One rule is meant to catch password spraying: one account failing to authenticate from three or more distinct source IPs within 10 minutes. The rule file, abridged, reads: ``` match: event = failed_logon group_by: src_ip threshold: 5 window: 10m ``` In production it fires about 144 times a day, and every alert comes from one address: a service account whose stored credential expired, retrying every 30 seconds against a single host. The domain's lockout threshold is 50 attempts, so no lockout events are produced. Which change makes the rule detect what its hypothesis describes?
MCQ-16A hunt for C2 beaconing flags hosts with at least 500 outbound connections to a single external IP over 30 days and a connection-interval coefficient of variation below 0.05. It returns 40 hosts. 38 of them run a signed monitoring agent that polls its cloud endpoint every 60 seconds. The other 2 poll a different external IP every 900 seconds, with intervals between 899 and 901 seconds. The team can triage 10 hosts this week. What is the correct next step?
MCQ-14A detection rule for suspicious service-account logons fires 1,200 times a day. One week of triage shows 1,150 of those alerts come from a vulnerability scanner's service account on 40 known hosts, each with a matching open change ticket. The remaining 50 come from the same account on hosts outside that list; 12 of those have no matching change ticket and no known owner. Two analysts can triage about 60 alerts a day. Which change should the team make first?
MCQ-12A detection team maps 140 rules to 96 ATT&CK techniques and reports that 78% of its in-scope techniques have a mapped rule. A purple-team exercise then runs 30 procedures against those mapped techniques: 19 produce no alert, 7 alert on the intended behavior, and 4 alert on unrelated benign activity in the same window. What does the exercise establish?
MCQ-17A 12-analyst tier-1 team triages ~900 alerts a day from phishing, EDR and cloud sources. The SOC manager wants a single headline number for the weekly ops review, and analyst ratings will be based on it alone. Which metric, used that way, most reliably rewards closing a ticket over resolving it?
MCQ-1503:40, one analyst on shift, five unassigned alerts in the queue. Which do you pick up first? - `pay-db-01`: production Postgres in PCI scope. EDR flags an unusual process spawned by `postgres`; `pg_dump` wrote to `/tmp` and a 1.8 GB outbound transfer to 203.0.113.44 completed over 12 minutes. That IP has no prior traffic in the environment and no backup job runs at this hour. - `dc-01`: domain controller. 40 RC4 TGS requests for service accounts in 90 seconds from `ws-4471`, logged in as a helpdesk account with no admin rights. No ticket crack observed. - `ws-2210`: marketing laptop. Mimikatz signature, blocked and quarantined before execution, no network connection. - `build-07`: CI server. One DNS query to a domain added to a threat feed 6 hours ago; the response was NXDOMAIN and no connection followed. - `vpn-01`: impossible-travel login, MFA satisfied, session ended after 30 seconds with no post-auth activity.