
Starting with just the company name, Secure.com uncovered an exposed fraud detection model and a cache control anyone could use. The team fixed both within days. A retest confirmed every reported issue was closed.
Ad-fraud protection
Infrastructure lead and cloud developer
One-day external penetration test
The startup helps performance marketers tell human visitors from bots. The signals its model uses to make that decision are among its most valuable assets.
On 3 June 2026, Secure.com mapped 31 domains and subdomains, eight public IP addresses and two cloud storage buckets. Of nine confirmed findings, two stood out. Anyone on the internet could reach the model files or interfere with how the company’s sites served visitors.
An infrastructure lead and a developer shared responsibility for security alongside their day jobs. After an investor introduction in May, the CEO accepted a free external pentest. The engineers wanted evidence they could check and a clear answer on what to fix first.
The scope was the public internet. Secure.com received the company name, with no credentials, VPN access or hints about where to look. The team ruled out denial-of-service testing and agreed to stop before any service disruption.
Over one day, Secure.com mapped the public assets, reviewed the application and cloud environment, validated suspected weaknesses, and assessed how individual findings fit together. The report, delivered on 4 June, distinguished what the team could demonstrate from what remained uncertain.
Of five potential attack chains assessed, one was confirmed end to end. Two were partially confirmed, one failed, and one was ruled out.
A development storage bucket exposed model files and the feature definitions that described how the platform distinguished bots from people. The bucket held development and demo content, but the detection logic itself was core intellectual property. Someone studying those signals could learn what the product looked for.
Clearing cached pages was meant to be an administrator action. Testing confirmed an unauthenticated visitor could trigger it across the main sites. Repeated clearing could send more traffic to the origin servers, increasing load and cost and slowing pages for visitors. The team verified the control failure without causing an outage.
During the 10 June walkthrough, Secure.com showed the feature names from the company’s own model. The infrastructure lead stopped the presentation to ask whether someone internally had pointed the testers to those files. No one had. The testing agent had located them from the public internet, starting with the company name.
The walkthrough became a working session. The developer identified which storage folders were exposed. The infrastructure lead checked the cache control live and confirmed the access problem. By the end of the call, both engineers knew what to fix first and why.
The engineers closed public access to the exposed model materials and restricted cache clearing to authorized requests. They fixed both priorities within days, then worked through the remaining issues using the report’s remediation roadmap.
Retest result. Secure.com verified that every reported issue had been resolved.