AI exposure: Penetration Testers

51.9% of this occupation's weighted task load is exposed, which puts Penetration Testers at the 90th percentile of 923 occupations. The capability is largely there. Its average task scores 3.1 out of 4 on what a current system can produce, and the frictions that hold other jobs in place are comparatively weak here.

What holds the line here is context. Across this occupation's 22 tasks it averages 1.91 out of 3, the highest of the five friction dimensions. In plain terms, the work depends on knowledge the model cannot hold. Much of this job runs on things that were never written down: what this particular organisation does, what happened last week, what the person across the table actually meant. That context is the barrier, and it erodes as systems are given more access.

The most exposed thing this job does is Gather cyber intelligence to identify vulnerabilities, at 93.3%. The least is Identify new threat tactics, techniques, or procedures used by cyber threat actors, at 10.0%. A gap of 83.3% between two parts of the same job is the reason this index publishes at task level. An occupation-wide number would have hidden both.

Within computer and mathematical occupations, this one is less exposed than the median of 57.8% across the group's 36 roles, with 25 scoring higher. Being in an exposed family does not make a particular job exposed, and the reverse holds too.

What would move this score. Of 22 tasks, 19 are currently banded exposed, 3 assisted and 0 untouched. For that distribution to shift materially would take cheaper ways to verify output, since the cost of checking is currently doing more to hold this work in place than the cost of producing it. The score is re-computed every quarter against a fresh capability reference, and the change is published rather than quietly applied.