Why the Incident Keeps Coming Back: A ResponseOps Diagnosis

Why the Incident Keeps Coming Back: A ResponseOps Diagnosis
Security Tech

A pattern repeats in security teams: the incident is “closed” on Friday, and by the following Wednesday the same alert is firing again. Leadership is frustrated, engineers are tired and someone suggests that the tools are faulty. More often the tools are fine and the response was incomplete. The Ops Atlas, in its entry on ResponseOps (the Atlas’s own nickname for keeping incident handling permanently ready), catalogues the usual reasons clean-ups fail. This article treats the recurring incident as a diagnosis problem: symptoms, likely causes, tests and fixes.

One clarification: no formal standard adopts the ResponseOps name, a point the Atlas concedes openly. Agreements should say incident response.

Symptom: the same intruder keeps coming back

Re-entry is the most visible failure. The Atlas lists it as the typical outcome when eviction happens before the scoping work is done. If you removed one backdoor while another stayed in place, the actor has no reason to start from scratch.

Likely causes

  • Incomplete scope. Persistent access mechanisms such as extra accounts, scheduled tasks, stolen tokens or remote tools were never inventoried.
  • Staged removal. Teams removed access piece by piece over several days, giving the intruder time to adapt.
  • Single-pass credential resets. Per the Atlas, a compromised central directory service calls for a specific key to be rotated two times, separated by about ten hours, since otherwise older copies of it still work.

Tests

Ask for the written list of every persistence mechanism found, and who verified it. Check that the list was completed before any removal began. Review whether all session tokens and cloud keys were revoked, not merely passwords, since temporary cloud credentials stay valid until their own expiry.

Fix

Pause, finish scoping and plan one coordinated removal. The guidance the Atlas relies on says to account for all persistent access before eradication. Then execute it as a single operation with a checklist and a witness.

Symptom: ransomware returns after restoring

Likely causes

The restoration source was itself compromised, or restored machines were reconnected before the intruder’s access was removed. The Atlas lists restoring from tainted backups as a classic route to a second encryption event.

Tests

Find the date the intrusion began, using the earliest evidence rather than the first alert. Compare it with the date of the backup used. If the backup is later than the earliest activity, it is suspect.

Fix

Choose an older, verified copy, scan it in isolation, rebuild identity systems first and reconnect machines in stages. The Atlas also notes that the guidance says eradication and recovery can be executed simultaneously, which is useful when the business cannot wait, provided the removal list is complete.

Symptom: the team cannot say what was taken

Likely causes

Evidence was overwritten by a rushed reinstall, or logs were never captured. A wipe and reimage removes one foothold and also the story of what the intruder did.

Tests

Look for a custody log, hashed copies and a record of what was collected in what order. The Atlas, drawing on a long-standing technical RFC, expects the most fleeting data, such as memory, to be taken before disk contents, with timestamps in a single time standard and working copies kept separate from originals.

Fix

Build an evidence routine into the plan: a write-once store, a custody template and a short procedure for the first responder. Practise it.

Symptom: a reporting deadline was missed

Likely causes

No named person judged the moment of organisational awareness, or the clocks had never been charted. The Atlas lists several regimes with different start points and lengths, for example a window counted from awareness with a standard of reasonable certainty in one regime, and a window counted from a materiality decision in another. This is a legal area, so consult a qualified adviser rather than relying on any summary.

Tests

Ask who in your organisation can say, in writing, that awareness occurred. If the answer is unclear, that is the gap.

Fix

Name the owner and a deputy, build the clocks map with counsel and rehearse the decision at least once a year.

Symptom: the response team’s own messages leaked

Likely causes

Response discussions took place in email or chat that the intruder could still read. The Atlas lists this as a recurring anti-pattern.

Fix

Maintain an out-of-band bridge, tested quarterly, and a set of emergency accounts that live outside your normal identity system.

Symptom: everything slowed to a crawl after containment

Likely causes

Containment was applied everywhere, immediately, including to systems the business depends on. The Atlas warns that sealing off every system in one go may alert the intruder or set off destructive routines, and recommends time-boxing any deliberate delay while weighing damage potential, evidence, availability, effort, effectiveness and duration.

Fix

Write containment options in tiers, from light to heavy, with a note of the business cost of each and who may authorise it.

Worked example: reading the timeline

Consider a made-up case. A file server shows odd activity on a Monday evening. The first alert arrives Tuesday morning, and the team spends two days cleaning up. Then the alert returns on Friday. When the team builds a proper timeline, they find the earliest intruder action was the previous Thursday, so the backup they trusted, dated the Saturday before the alert, was already tainted. They also find a scheduled task on a second server that was never listed.

Each finding maps neatly to a symptom above: the backup date test, and the persistence inventory. Notice that none of this needed new tooling. It needed a timeline anchored to the earliest evidence, which is why the Atlas separates dwell time (from first intruder action to detection) from the time the team took to react. The first measures how long you were blind; the second measures how well you responded once you could see.

Preventing a repeat

Once the immediate fix is done, convert each diagnosis into a standing control. Add a mandatory persistence inventory to the plan. Require a backup date check before any restoration. Include an evidence capture step on the first responder card. Create the awareness owner role. Test the side channel every quarter. Tier the containment options. Together these six controls remove the most frequent reasons that incidents refuse to stay closed, and they fit on a single page.

A diagnostic table in words

  1. Same alert returns: check scope completeness and credential resets.
  2. Encryption returns: check backup dates and reconnection order.
  3. No answer on data loss: check evidence capture.
  4. Late notice: check awareness ownership and clocks.
  5. Leaked plans: check the side channel.
  6. Business paralysis: check containment tiers.

Giving the team room to reset

After a long recurring incident, people need a break. Planning a short trip can help, whether it is exploring some friendly routes beyond the usual Gujarat sights or reading a first-timer’s guide to what to do across the state. A rested team diagnoses better.

Further reading

For a fuller catalogue of failure modes, metrics and maturity stages, consult the Atlas page on ResponseOps in full. Use it as a checklist for your next post-incident review, and write down which of the symptoms above you recognise.