Security
Vulnerability research: most of the work is ruling things out
A recon pass hands you hundreds of finding-shaped candidates. Which ones die is a person's call; two gates around that call are mechanical.
Computing America Cyber. Security assessments, penetration testing and staff training.
Name the component
In short
- A candidate list is not a findings list. Ask any testing firm what it discards and on which test, because what reaches you is whatever survived.
- Treat a scope document as a technical control. Our runner refuses to send a packet until an authorization document names the host, because the alternative is a judgment call made at two in the morning.
- A mechanical validator can enforce that a draft is complete, has a well-formed CVSS vector whose score matches its metrics and carries a captured request. It can't decide whether a finding is real.
- An unreproduced report is expensive to disprove: curl's confirmed-report rate fell from above 15% to below 5% during 2025, against rising AI-generated submissions.
In our own testing, most candidates that look like findings get discarded before anything is written up. In one recorded pass our harness surfaced 253 takeover candidates, and none was real.
If you buy security testing, what you receive is whatever survived that discard, so the discard is the work you're paying for. Between client engagements we hunt on public disclosure programs: companies that have published a policy inviting outside testing of named assets, on stated terms. We get no credentials and the program owes us no agreement. A claim counts only once it reproduces in front of a stranger who would rather it didn't.
The gate before the first packet¶
The runner treats the scope document as a technical control. With no authorization document naming the hosts, it doesn't run in a reduced mode or print a warning. It writes a template to fill from the program's policy page, and refuses. An unfilled template parses to zero in-scope entries, so the refusal holds until someone pastes in the program's own words. Authorization is a legal boundary, and it gets tested when an interesting hostname resolves at the edge of scope, late, when you're tired. The tool enforces the rule at that moment.
# This script only performs reconnaissance, crawling, templated known-issue checks,# and GET-based content discovery. It does not submit forms, upload files, brute# force credentials, exploit findings, or mutate application state. With --verify it# additionally runs read-only active probes (canary reflection, open-redirect# sentinel, hidden-param discovery, SSRF OOB probe prep) plus, like the GraphQL# stage, a non-mutating introspection query; still no state-changing requests.Almost every candidate is not a finding¶
A recon pass over a mid-sized estate produces hundreds of candidates: hostnames pointing at deprovisioned cloud resources, exposed source maps, forgotten staging origins, endpoints that answer without a session, dependencies with a published advisory. Each is finding-shaped. Almost none are findings, and each dies on a specific test:
- A dangling DNS record is a takeover only if the service it points at will let you claim the name today.
- An exposed source map matters if it reveals internal API endpoints or a key, not if it reveals that the front end is TypeScript.
- An unauthenticated endpoint that returns data matters only if the data isn't already public; one that changes state matters regardless.
- A dependency advisory applies only if your code reaches the vulnerable path.
A researcher checks each candidate against the target's threat model before anything is written up. A report you can't reproduce costs your engineers time to disprove, and it costs the sender credibility on every later finding.
Group: Mechanical gates
Scope document
Mechanical: no document, no packets
- To Recon candidates
Recon candidates
Hundreds per pass; one recorded pass: 253 takeover candidates, none real
- To Threat-model call · most discarded
Group: A person decides
Threat-model call
A person: does it matter to this target? Most candidates are discarded here
- To Reproduced by hand · survivors
Group: A person decides
Reproduced by hand
A person captures the request
- To Mechanical validator · drafts
Group: Mechanical gates
Mechanical validator
Mechanical: complete, valid CVSS, captured request
Two mechanical gates bracket a person's threat-model call, which discards most of the hundreds of candidates each pass produces.
SourceOur harness's process.
What a validator can and cannot enforce¶
The second mechanical gate is the draft validator. Before a draft leaves the machine it's linted against the report anatomy. That means a title, summary, severity, affected asset, numbered reproduction steps, impact and remediation, and a proof section carrying the captured request or response itself (a description of one doesn't count). The severity must carry a well-formed CVSS vector whose score matches its metrics. Leftover placeholders fail it, and so does an unredacted secret. A proof section that is only prose goes to a human.
Four questions decide whether a candidate survives. Is the triggering party untrusted across a real boundary? Does the attack need a configuration real deployments don't use? Has the affected code shipped, and who owns the root cause, the application or a dependency? Is the asset in scope for a reward, or only for reporting? Each of those questions has closed a technically correct finding in our corpus, and a linter can't answer any of them.
We use language models heavily for the first half of the work and not for the claim itself. They're good at breadth: enumerating a surface, correlating one host's behavior against another, and reading a lot of source. They can also write fluent reports about vulnerabilities that don't exist.
curl, which ran a bounty from April 2019, saw a confirmed-report rate that had run above 15% for years fall below 5% during 2025 (opens in a new tab), against rising AI-generated submissions that sometimes took a long time to debunk, and ended the bounty on January 31, 2026.
Ask a testing firm these four things¶
- 1.What stops your tooling at the edge of my scope? The answer should be a control that refuses to run without a scope document, not a promise to be careful.
- 2.What did you discard on your last engagement, and on which test? A firm that can't describe its discards hasn't ruled much out.
- 3.Does every report carry the captured request or response, and not a description of one?
- 4.Who reproduced each finding by hand before it reached me, and who decided the asset was in scope?
Sources
- 1.The end of the curl bug-bounty (opens in a new tab), Daniel Stenberg (curl),
Next step
Name the component you cannot get assurance on.
We will tell you whether it needs live testing against the running system or a review of its code and configuration, and what the first day would look for.
- Phone
- (214) 723-2510
- Reply
- A person replies, not a sequence: within one business day, from someone who would be on the engagement.
Packages
Start with a package
Four packages, each with its price on the page. Pick one and buy it online.
$199/month
Billed monthly
$182.42/month
Billed yearly
$499/month
Billed monthly
$457.42/month
Billed yearly
- Scammers stopped from sending email as you
- Your website and systems scanned every week
- Your email settings checked every week
- Everything in the 1 to 10 plan
- Covers up to 50 people
$199/month
after a one-time setup fee
- A site planned, designed and built for you
- Hosting and updates every month
$499/month
Billed monthly
$457.42/month
Billed yearly
$999/month
Billed monthly
$915.75/month
Billed yearly
- Your Google profile, fully set up
- A Google post every week
- 2 social posts a week
- Everything in Get found
- 4 social posts a week
- Your top 10 Google searches tracked
from$500/month
after a one-time setup fee
- Your data gathered from the tools you use
- A first analysis that answers your questions
- New findings every month
Related reading
Security3 min readPenetration test vs vulnerability scan: what you are actually buying
Both are legitimate purchases. A sample report and a look for chaining show which one a quote is selling, and four questions test the vendor behind it.
Security4 min readVulnerability prioritization: you do not have 4,000 critical CVEs
A CVSS-sorted export doesn't say what to fix first. CISA's June 2026 directive shows what to sort by instead, and a manufacturer or clinic can borrow it.
Security5 min readOT network anomaly detection with fuzzy hashing (TLSH)
How a TLSH similarity hash over Modbus/TCP packet clusters flags drift, the alternatives that failed, and what the method misses.
