AWS Security Agent can now hard-cap autonomous pentest spend and revalidate individual fixes

On August 19, 2026, AWS added two operational controls to AWS Security Agent, now part of AWS Continuum. Penetration tests can be created with a maximum billable task-hours limit using a preset, custom value or no limit; when cumulative agent work reaches the ceiling, the test stops gracefully and preserves findings discovered so far. Teams can also select findings from a completed test and revalidate only those issues against the live application after deploying a fix, receiving an Active or Resolved result with revalidation history tied to the original finding. AWS prices autonomous penetration testing at $50 per task-hour, metered per second, and task-hours represent cumulative parallel agent work rather than elapsed test duration.

Autonomous security testing creates a particularly awkward usage-pricing risk because the agent decides how much exploration is needed and can execute several tasks concurrently. A four-hour pentest can therefore consume far more than four billable task-hours. The new ceiling turns that uncertain exploration budget into an enforceable run-level boundary instead of relying only on monitoring or manually stopping the job. Finding-level revalidation changes the post-fix workflow as well: teams can verify a remediation without paying for and waiting on another full application test. The trade-off is coverage—setting a ceiling too low can stop a test before the agent finishes exploring the intended attack surface.

AWS bills penetration testing by cumulative task-hours. If the agent performs multiple tasks in parallel, those task-hours add together, so billable usage can exceed wall-clock duration. AWS’s own pricing example shows a four-hour test consuming 24 task-hours, which would cost $1,200 at the current $50 rate. This makes an execution ceiling materially different from an ordinary elapsed-time timeout.

A penetration test can now specify `maxTaskHours`. AWS supports preset or custom limits, and when cumulative usage reaches the configured maximum the job is gracefully stopped. Findings already discovered remain available. A higher ceiling does not itself cost more unless the agent actually consumes the extra task-hours, so teams can choose a risk budget that reflects the application and test scope rather than purchasing capacity upfront.

Completed findings can now be selected for a targeted revalidation run. AWS Security Agent retests the specific vulnerability against the live application and records whether it remains Active or is Resolved, with the revalidation history linked back to the original issue. AWS says the first three revalidations per finding are currently free; additional revalidation work is charged by task-hour.

A task-hour ceiling protects spend, but it does not tell the agent which parts of an application matter most. AWS recommends narrowing URL scope and selecting only the relevant risk types to reduce work. Teams should treat the budget as a last-line guardrail and use representative previous runs to choose a realistic limit; there is no precise pre-run cost estimator because the agent adapts its testing plan as it explores the application.

Gracefully stopping at a cost limit preserves useful findings, but it does not mean the intended test coverage was completed. Security teams should record whether a run stopped because testing finished or because the budget was exhausted, and avoid presenting a capped partial run as equivalent to a completed penetration test. Where coverage is mandatory, the response to an exhausted budget may be to narrow scope, raise the ceiling deliberately or complete the work with another testing path.