Researchers and start-ups working on quantum computing mostly reach their machines through the cloud, and their data, notebooks and experiment results sit in ordinary cloud accounts. That means the unglamorous discipline of CloudSecOps applies to them as much as to any web shop. Emerging-technology circles also breed confident folklore about security, so below are six popular beliefs, each checked against the Ops Atlas, the editorial glossary kept by xOps School, whose entry on the subject is the source for the facts below.
Myth 1: “Our cloud vendor has security covered”
Reality. Providers protect their facilities and core platform. Account settings, identities and stored data stay with the customer whichever service model is bought, per the Atlas. A research group that leaves a results bucket open is not protected by the size of its provider. If your experiments contain unpublished work or partner data, treat the settings as part of the lab’s equipment, to be checked on a schedule.
Myth 2: “Rotating our encryption keys protects the data”
Reality. Rotation is good hygiene, but the Atlas cites AWS documentation on a pointed detail: rotating a key does not change the protection of data the key already guards. The safeguards that matter more are narrow access to the key, logs of its use and a deletion process with a waiting period. The Atlas lists windows of seven to thirty days on AWS with a default of thirty, seven to ninety on Azure’s vault with purge protection, and thirty on Google’s key service. Use them; a key scheduled for deletion can still be cancelled while the data behind it is intact.
Myth 3: “Our threat detection service keeps all our logs”
Reality. One widely used AWS detection service extracts fields from log data and then discards the raw records, according to the Atlas. Detection and retention are different jobs. Anyone who expects to reconstruct a quarter-old incident from a detection console will find gaps. The Atlas recommends organisation-wide logging into an archive workload that normal administrators cannot change, and it notes that the default look-back on major platforms is often ninety days, with detailed data events off until enabled.
Myth 4: “We are fully compliant with the CIS benchmark”
Reality. That sentence is incomplete without an edition and a method. The benchmarks keep advancing, and the Atlas points out that one provider’s own posture checks lag behind the latest editions; it also notes that some benchmark items are manual and cannot be tested by a scanner at all. A precise statement names the version, lists the manual checks performed and records the exceptions with their expiry dates. Any quantum team presenting results to a funder would insist on the equivalent precision for an experiment, so apply the same standard here.
Myth 5: “A high pass rate means we are safe”
Reality. Pass rates can be gamed. The Atlas observes that in AWS a control may be reported as passed when its findings were suppressed rather than repaired. A tidier gauge is coverage plus dwell time: the share of accounts under current rules, and how long serious findings remain open before closure, reported with the median and the long tail. In the Atlas’s illustration the typical closure takes eight hours while the arithmetic mean passes thirty, a gap that shows how a few slow items hide inside an average.
If you enjoy statistics, the logic will feel familiar. An average conceals the distribution, just as a single measurement hides the noise in a quantum run. For a sense of how algorithms handle searching large spaces of options, see our explainer on the way quantum algorithms approach tough optimisation tasks; triaging a long list of findings is a humbler optimisation problem, but the instinct to rank by consequence is the same.
Myth 6: “We revoked the user, so the access is gone”
Reality. Revocation is often incomplete. Temporary credentials issued by one major provider can outlive a working day by many hours, and blocking by token issue time cannot reach a session that someone assumed afterwards. The Atlas advises removing the path that allows a role to be assumed again before attempting revocation. Rehearse this in a drill, not during an emergency.
A note on how myths spread
Folklore thrives where a topic is new and vocabulary is loose. The Atlas is frank that CloudSecOps is a label coined in trade coverage, not a standard, and that one analyst called it a buzzword when it first circulated. That candour is useful for readers in a fast-moving field: if a term has no formal definition, a vendor can stretch it to fit a product, and a conference talk can stretch it further. Ask each time what is actually being checked, by whom and how often. A belief that survives that question is probably sound; one that dissolves is a myth.
The same habit helps when reading research claims in quantum computing itself. Look for the benchmark, the conditions and the caveats, and be wary of a headline figure without them. Security folklore deserves no kinder reading than a press release about qubits.
What the true basics look like
The list above should leave you with a short working routine, none of which needs special hardware. Know which accounts exist and who owns them. Keep logs in a protected archive. Hand every serious finding to a named owner inside one working day, and close it only after a rescan. Cap exceptions at ninety days. Trial blunt rules in a test organisational unit first, since the Atlas lists deny-everything-on-day-one among its anti-patterns.
For a fuller treatment of scope, including which topics the Atlas leaves to neighbouring disciplines, read the CloudSecOps page in the Ops Atlas. The entry is honest that the term is emerging and that some of its structure is editorial rather than standardised.
Where emerging-tech teams fit in
Quantum computing may change what attackers and defenders can do with cryptography over time, a subject for specialist writing. Today, the more probable cause of an exposed research dataset is a plain misconfiguration. If your group combines classical and quantum workflows, as discussed in our piece on how quantum methods are reshaping machine learning, the classical cloud layer deserves its share of attention.
A last note on proportion. If your team is two people with one cloud account, the Atlas, quoting Microsoft’s guidance, suggests a combined role rather than a dedicated function. Do the basics, write them down and revisit them each quarter. Myths fade when each belief is replaced by a check that someone actually runs.