Continuous AI VAPT vs Annual Pentest: Why Agentic Attackers Changed the Cadence

Why annual was ever the number

Pentesting cadence was always a compromise between risk and cost. A senior pentester for two weeks against a mid-sized application cost on the order of $30–80k, and the output was a point-in-time report. Doing that quarterly was prohibitive for most companies; annually was tolerable; biannually was the compliance minimum. The rhythm matched the attacker's rhythm: a determined human attacker also cost real time and money, so the expected interval between serious attack attempts was itself months. The yearly cycle was defensible.

What agentic attackers changed

An offensive framework like ARTEX reduces the per-target cost of a competent attack attempt by at least an order of magnitude, probably two. Three things follow directly:

• The set of "targets worth attacking" expanded. Internal sales tools, developer-facing dashboards, documentation portals, AI gateways, status pages — all previously below the attacker's ROI threshold — are now routinely swept.
• The cadence of attempted attacks increased. Attackers do not wait six months to try again; they run the framework daily against the surfaces that moved.
• The first-move time dropped. New code shipped at 14:00 is probed by someone's agent before midnight. Mean time-to-first-probe is measured in hours, not quarters.

The defender who still runs an annual pentest is not wrong, but they are certifying a surface that looked a certain way eleven and a half months ago. The attacker who ran yesterday saw a different one.

What "continuous AI VAPT" actually means

The term is doing a lot of work. In practice it has five properties:

1. Agent-driven. The probe engine chooses follow-up actions based on target responses — same class of tool the attacker is running against you, operated under consent.
2. Scheduled. Daily or on-change, not quarterly. The point is to compress the gap between "surface changed" and "surface tested."
3. Scoped by inventory, not by human memory. The engine tests the live attack surface pulled from your CMDB, cloud accounts and attack-surface map — not the list the last pentester happened to be told about.
4. Deduplicated and prioritised. Daily testing produces daily volume; a human-usable program needs findings grouped by root cause, enriched with exploitability and reachability, and ranked against business impact.
5. Evidence-producing. Each finding has a reproducible payload and a safe-to-replay artefact, so a developer can confirm the fix without re-engaging a pentester.

What continuous VAPT does not do

An honest picture matters.

• It does not replace a human looking for business-logic bugs — the "can I check out with a negative quantity?" class of defect that requires understanding what your product means, not just what it exposes.
• It does not replace the adversarial creativity of a senior tester who's seen a specific pattern fail three times in prior engagements.
• It does not satisfy every compliance regime by itself. SOC 2, PCI DSS and ISO 27001 still expect a human-signed pentest report on cycle — though the cycle length is softening as auditors accept continuous programs as compensating evidence.
• It does not replace red-team exercises for high-value scenarios (social engineering at scale, physical access, insider threats). Those are different categories.

The right stack

The continuous layer is the floor. Everything above it is better, not a substitute. The failure mode is skipping the floor because a human pentest is on the calendar.

What good looks like in a continuous VAPT program

• Surface sync. The engine pulls the test surface from your ASM inventory daily; new assets are tested within 24 hours of appearing.
• Blast-radius guardrails. Payloads are safe-by-default, rate-limited, and bound to pre-approved CIDRs. The engine refuses to probe production without explicit scope grant; it will not touch customer data stores by policy.
• Ticket-first delivery. Findings land as deduplicated tickets in Jira / Linear / GitHub with reproducible payloads, affected endpoints, suggested fix, and a one-click re-test. No PDFs.
• Attack-path linking. Each finding is tied to an attack path — "this XSS plus this CORS misconfig plus this leaked key gets to X" — so prioritisation is about business impact, not CVSS.
• Comparable over time. Week-on-week trend on open findings, MTTR, and newly introduced issues per deploy.
• Evidence for compliance. Signed artefacts of every run, retained for your SOC 2 / ISO 27001 cycle, so the "annual pentest" conversation becomes "here's the continuous evidence plus the scoped human test on top."

The economics

An annual pentest of a mid-sized application costs $40–80k and yields a point-in-time report. A continuous agentic VAPT program costs on the order of a mid-range SaaS subscription and yields a live feed. They are not substitutes: the human engagement catches a different class of issue. They are stackable. The programs that will look obvious in 2028 run both; the programs that will look like the 2015 of this story only run one.

How TigerGate helps

• TigerGate's VAPT AI Engine runs an agentic offensive-security workflow against your apps on a schedule — the same class of tool ARTEX and its peers run for the other side, operated under your consent. See continuous VAPT for the full program.
• Attack Surface Management feeds the VAPT engine with a live inventory of internet-exposed assets, so coverage follows reality rather than last quarter's spreadsheet.
• Attack Path Analysis joins findings into chains and tells you which ones actually reach a business-critical asset.
• DAST and API Security run alongside, covering the known-class issues the agentic layer uses as its baseline.
• Compliance evidence packages continuous VAPT results as the evidence your SOC 2 / ISO 27001 / PCI auditor will accept alongside your scoped annual test.