Aikido launches external network pentesting

Aikido launched external network pentesting, a service that uses AI agents to map and attack a customer's internet-facing systems starting from only a domain name, the company announced. Aikido pentesting lead Wout Debaenst framed the offering as running the same test attackers already perform: "You can let us do the hacking, or you can let criminal groups go at it." Aikido cited Cloudflare data showing thousands of exploit attempts against the PHP2Shell PHP flaws in the 10 hours before public disclosure, rising to millions by the next day, and a16z-charted Zero Day Clock figures showing 86.7% of exploited vulnerabilities in 2026 were hit on or before disclosure day versus 18.7% in 2018.

TL;DR: Aikido now pentests your external network . Attackers already test your perimeter every day, and with AI agents doing the grunt work, they're faster than ever. Now you can run the same test first. Give our agents one domain and they'll map the internet-facing systems you run and attack them from the outside, exactly like a malicious actor would. Somewhere right now, automated tooling is working through your subdomains, fingerprinting your servers and checking them against known vulnerabilities like PHP2Shell https://www.aikido.dev/blog/critical-vulnerabilities-in-php-php2shell , the critical PHP flaws that went from leak to active exploitation in a couple of hours last week. AI agents can now do more than just spot an outdated framework or an exposed config file. They’ll try the exploit, adapt when it fails and use whatever they find to reach the next system. Weaknesses that used to take a skilled human and a free weekend have become low-hanging fruit that gets picked automatically, across thousands of companies at once. How fast is that? When PHP2Shell went public, Cloudflare had already seen thousands of exploit attempts https://www.channelnewsasia.com/world/cloudfare-hack-exploit-attempts-php-websites-6481421 in the 10 hours before the announcement. That jumped to tens of thousands within 45 minutes, hundreds of thousands within two hours and millions by the next day. At that speed, the official response to the worst flaws can be to pull exposed systems offline until they're fixed. When attackers went after Ivanti's VPN appliances in 2024, the US Cybersecurity and Infrastructure Security Agency CISA gave federal agencies 48 hours https://techcrunch.com/2024/02/01/cisa-federal-agencies-disconnect-ivanti-vpn to disconnect every affected device, and only let them back online once they were rebuilt and fully patched. And PHP2Shell isn't an outlier. Zero Day Clock numbers charted by a16z https://x.com/a16z/status/2107565314410230224 show that in 2018, 18.7% of the vulnerabilities attackers exploited were hit on or before the day they went public. So far in 2026, it's 86.7%. But you can only pull the plug on a system you know about, and only after someone tells you it's at risk. When attackers move this fast, the only way to get ahead of them is to test just as often. Wout Debaenst, Aikido's pentesting lead, has spent years running these tests by hand: "What we're testing is exactly what's already being tested constantly. You can let us do the hacking, or you can let criminal groups go at it. It's the same test and you'd rather have us find it than them." Starting today, you can run that test first. Aikido's external network pentesting attacks your perimeter the way an outside attacker would, starting from nothing but your domain and a small army of AI agents. What is external network pentesting? Pentesting terms get used loosely, so here's what each one usually means: - Web app pentest: goes deep into one application, its logic, its user roles and its code. This is where business logic flaws and broken access controls show up. - External network pentest: covers everything reachable from the public internet under your domain, like servers, VPN gateways and mail servers. Your apps are in scope too, but only the part anyone on the internet can see: the server they run on and the pages you reach without logging in. Testing what happens after login, like user roles and business logic, needs a web app pentest. - Internal pentest: covers systems that aren't reachable from the internet, like an office network, a private cloud or Active Directory. External network pentesting doesn't include these. If most of your business runs on one application, start there. Run a web app pentest https://www.aikido.dev/attack/aipentest on that application first and fix what it finds. Then run an external network pentest to test everything around it, so the rest of your perimeter doesn't give attackers an easier way in. Your perimeter changes faster than your pentest schedule Many companies pentest their external network once or twice a year, and anything that changes after the testers finish goes untested until next time. If someone spins up a new subdomain the week after the pentest, it can sit there for months before anyone on your side tries to break into it. In Aikido's 2026 State of AI in Pentesting report https://www.aikido.dev/state-of-ai-pentesting , 48% of the 400 security and engineering leaders we surveyed said pentest findings are already outdated by the time they arrive. So you need to test more often, ideally every time your perimeter changes. Run continuously, Aikido's external network pentest keeps mapping everything under your domain, and agents start testing each new host shortly after it goes live. Continuous pentesting is priced yearly, based on the size of your external network at the time of the first pentest. If you're still working out how often to test what, here's how to plan your pentest frequency and budget around risk https://www.aikido.dev/blog/more-pentesting-same-budget-ai . How it works 1. Agents map your perimeter. They search for subdomains and internet-facing systems under your root domain: websites, apps, APIs, VPN gateways, mail servers and self-hosted tools like a GitLab server. Subdomains pointing to services you don't own get filtered out automatically, and you can remove any others you don't want tested. 2. You verify your ownership. Once it's confirmed, you get an exact quote and launch the pentest. There's no scoping call, and you don't need an existing Aikido subscription. 3. Agents attack each host. Every host gets its own agents, which work through the server itself and try to get in with known CVEs, misconfigurations and exposed secrets. 4. Only proven findings make the report. Before findings are accepted, agents need to back up their claims with concrete evidence that is validated and reproduced by other agents. 5. You fix and retest. Each finding comes with a fix suggestion. Once you've applied it, a retest spins up an additional agent that checks whether the issue is gone, and whether the fix created a new way in somewhere else. Retests are available for up to 6 months. Agents that decide their next move Aikido's pentest is built around agents. Each one starts from a different point with its own focus, and decides what to do next based on what it finds. A version banner, an open port or a leaked file changes the plan, the same way it would for a human attacker. That's what lets them adapt to your environment, however it's set up. Most automated pentesting tools, including some marketed as AI, work from a precoded set of checks. If a check doesn't match your environment exactly, that part goes untested, and that's the kind of blind spot an attacker finds right away. Here's what adapting looks like on a typical perimeter: 1. An exposed .git folder makes repository history readable on a public server 2. The repository history contains SSH credentials 3. The same credentials work on a different host 4. On that second server, the agents find Azure credentials 5. Those credentials give admin access to the cloud tenant Few of those findings would top a report on their own. Chained together, they hand over the cloud tenant. The report shows the full chain, so you can see what an attacker could actually reach and how each step got them there. A tool for your pentest team Human testers are at their best on the deep, creative work, like a business-critical app or an attack scenario only someone who knows your business would think of. Their time is worth the least when it goes into re-checking hundreds of hosts for issues that showed up since the last engagement. AI agents can and should start doing that part. They cover every host they find on each run, so the people you pay for spend their time where judgment matters most. For pentest vendors, the idea is the same: put the agents across a client's perimeter first, then spend the engagement on the work that needs your attention. Safe to run on production Your external network is your production environment, so the agents are built to test it without affecting anything: - A supervisor agent reviews every action before it runs and blocks anything that could disrupt systems. - Agents stop once exploitation is proven and ask for human confirmation before going deeper. - They don't download or store your data. Findings show the access the agents got, without copying anything. - Beyond your root domain, ownership of every individual target gets checked, so agents never test systems that aren't yours. - The panic button stops every agent instantly. Get started Watch the demo to see the agents work through a perimeter. If you'd rather jump straight into your own, then simply pick the domain you care about most, and let the agents go to work. You can follow every agent live while it works, and the report is ready within hours. {{cta}}