Penetration Testing / Artificial Intelligence TL;DR Attackers move first. On average, exploitation now starts before a patch exists (Mandiant M-Trends 2026). In 1H 2026, 23% of vulnerabilities VulnCheck added to its exploited list were hit on or before disclosure day. AI pentesting vendors aren't interchangeable. Most look convincing in a demo. The gaps show up in your logged-in flows, your coverage, and your production safety. Make them prove it on your own web apps. Score each vendor on context, reach, validated findings, guardrails, and retests. Choosing an AI pentesting vendor? Get 15 questions, red flags, and a scorecard to test what vendors deliver, not just what they demo. Download the free checklist → AI is speeding attackers up For the first time, Google's Threat Intelligence Group identified a threat actor using a zero-day exploit it believes was developed with AI: a 2FA bypass rooted in a logic flaw that conventional fuzzing and static analysis can easily miss (GTIG). Agentic pentesting can close that gap, and 87% of US security leaders surveyed are planning, piloting, or using it. Choosing the agent is the hard part. Every Demo Looks Good. Your App Is the Real Test. Picture three agents that all demo well. The first finds real flaws but stops at your login page. The second navigates your app but can't answer a basic question: coverage of what? The third exploits real issues but has no hard limits for production. Each looks impressive on the vendor's target. The difference shows up on yours. A structured bake-off surfaces it before you sign. Ask What the Agent Knows Before It Attacks Every tester has to learn your site before the first payload. That recon is real attack work. The cost is redoing it from scratch on every engagement, when most of the map hasn't changed. In CyCognito's model, recon and analysis take 20 to 30 of a manual pentest's 63 to 95 hours. Agents differ here. Some start from a blank URL and rebuild the map on every run. Others start from an existing model of your site and keep it current between runs. Ask which one you're buying, what carries over between runs, and whether the agent keeps discovering beyond its starting map. A starting map should be a head start, not a boundary. Without context, an AI agent behaves like a smart scanner. With context, it can reason about your application. That changes what it finds, not just how fast. When a security researcher hunted for flaws in Apache Airflow, their AI assistant called the search a dead end. Testing how the deployment’s parts trust each other, instead of reading the code, turned up an unauthenticated RCE (CVE-2026-68982). Ask for Proof, Not Suspicion A suspicious pattern is not an exploitable vulnerability. Ask whether the agent exploits what it suspects, reproduces its findings, and hands over evidence your team can verify. A list of suspected issues is not a validated pentest. Here is what proof looks like. During an authorized assessment of a major insurer's production website, an AI pentesting agent found that the one-time passcode request could send login codes to any email address. It proved the flaw by logging in with a code sent to its own inbox, and guardrails then stopped the test. The insurer said it had pentested the site about six months earlier (Reflectiz case study on The Hacker News). Three Questions That Expose a Weak AI Pentesting Vendor A demo can look convincing without proving that the agent reaches your logged-in workflows, validates what it finds, or confirms your fixes. Here are three of the 15 checks to run before you choose. The other 12, covering starting context, coverage proof, attack chaining, access across roles, data handling, and guardrails, are in the full checklist. Run a Bake-Off, Not a Demo Demos run on the vendor's target. Bake-offs run on yours. Same target, same week. Same app, test accounts, and scope for every vendor. Log three timestamps. Access granted, discovery complete, first test. Score every check. The checklist's scoring sheet weights what matters and gates on safety. What's Inside the Checklist 15 checks, each with the question to ask and the red flag Checks for attack chaining, roles, guardrails, and retesting A 36-point scorecard and a per-check scoring sheet The best AI pentesting agent isn't the one with the best demo. It's the one that can show what it tested, prove what it found, and run safely on your application. Use these 15 questions to find it. Download the free AI Pentesting Checklist and vendor scorecard → Found this article interesting? This article is a contributed piece from one of our valued partners. Follow us on Google News, Twitter and LinkedIn to read more exclusive content we post.
AI Pentesting Checklist for Web Apps: 15 Questions to Ask Before Choosing a Vendor
Full Article
Original Source
Read the full article at Thehackernews →KhanList aggregates and links to publicly available news content. We do not host full articles from third-party sources. Always verify important information with original sources.