AI Pentesting Is Spreading Fast, but Data Privacy and Production Safety Hold Leaders Back
Software is being built and shipped faster than ever, and attackers are moving just as quickly, with time-to-exploit now down to a single day. A survey of 500 security and engineering leaders in the US and UK, run by security vendor Intruder, shows how far testing has fallen behind. Some 68% of organisations deploy code to production at least daily, yet 47% go more than a month between full pentests of their web applications and APIs. Only 4% of leaders believe their current testing frequency gives sufficient coverage, and 42% accept the risk and ship without extra security checks in between.
AI pentesting is emerging as the answer to that gap. According to the survey, 55% of organisations already use it across their environment and a further 36% have tried it. Nine in ten leaders agree the traditional annual pentest is dead, and only 6% still rely on one a year. Over the next three to five years, 54% expect AI to handle the bulk of testing.
The main concern is not whether the tools are good enough. Only a quarter name quality as a top barrier. The biggest worries are data privacy (40%), safety on production systems (36%) and whether auditors and regulators will accept AI-generated reports (28%). Trust is also tied to evidence: 46% say a lack of visibility into what was tested, or a lack of reproducible proof that a flaw is exploitable, would damage their confidence most.
The survey also shows the cost of validating findings. More than half of teams spend six hours or more a week on it, and fewer than two in five retest a fix in production as soon as it ships. Intruder suggests starting in a staging environment that mirrors production, running an AI pentest alongside the next manual one, and checking how code and data are handled before granting access.
Why it matters
Teams that ship daily but test monthly are leaving new code unexamined for weeks, and the survey shows many are knowingly accepting that exposure. The assumption that tool quality is the hurdle is wrong: the real decision is how much access to give an automated tester on sensitive systems, and whether its output will stand up to auditors. Leaders also need evidence such as a record of what was tested and reproducible proof, or they will simply trade one validation burden for another.
What to do
- Run AI pentests first in a staging environment that mirrors production.
- Run an AI pentest alongside your next manual pentest and compare the results.
- Look for findings that include proof-of-concept evidence and a record of what was tested.
- Check how your code and data will be handled before giving an AI pentest tool access.