Skip to content
← Insights

AI Didn't Make Attackers Smarter. It Made Your Easy Stuff Cheaper to Find.

November 16, 2026 · 6 min read

AI Summary
Generating summary…

Everyone wants to talk about AI writing novel zero-days. That's the movie version.

The real version is boring. That's exactly why it'll hurt you.

In 26 years and more than 600 incident response engagements, I can count on one hand the breaches that started with something truly sophisticated. The rest started with a VPN that never got MFA. A firewall two firmware versions behind. A password that showed up in an infostealer log three months earlier, or a server nobody remembered owning. Low-hanging fruit. Every time.

AI doesn't change what that fruit is. It changes how fast and how cheaply someone can pick it.

The economics flipped

Attacking used to cost time. Someone had to read the advisory, understand the bug, write the exploit, test it, then go find targets. That cost acted as a filter. Attackers spent their hours on targets worth the effort, and a 40-person manufacturer in a small town mostly wasn't one.

That filter is gone. Cogent Security analyzed close to 70,000 CVEs and found the time attackers needed to build a working exploit dropped from about 125 days in January 2025 to half a day by April 2026. Half a day. Researchers have also shown LLM pipelines that turn an advisory and a patch into a working exploit in 10 to 15 minutes.

When building the exploit costs almost nothing, scanning the whole internet for it costs almost nothing too. Nobody is picking you. They're picking everybody, and you're on the list because you're on the internet.

"We're too small to be a target" was always a bad argument. Now it's just wrong.

Easy bugs are the ones AI loves

The vulnerabilities AI speeds up the most are the simple ones.

Rapid7's latest quarterly report found 79% of the high-severity flaws being exploited need no authentication and no user interaction. No phishing email. No click. Just a reachable service and a known bug. That's the class of thing a model can reason through fast and an automated pipeline can fire at a million IPs before lunch.

The 2026 Verizon DBIR now lists vulnerability exploitation as the most common way attackers get in. And once they're in, they don't hang around. CrowdStrike puts average eCrime breakout time at 29 minutes, with the fastest at 27 seconds.

The gap between exposed and owned is shrinking from both ends.

The usual suspects

When I walk into an incident, these are the doors I find propped open. None of them are new. All of them are now cheaper to find.

Edge devices. Firewalls, VPN concentrators, gateways. They sit on the internet by design, they often run old firmware, and they rarely have EDR. Qualys flags edge devices as carrying the highest strategic risk per vulnerability. They're also the boxes teams are most scared to reboot.

Remote access without MFA. RDP exposed to the internet. A VPN on password only. A vendor's support connection into your environment that nobody has reviewed since 2019. A credential is all an attacker needs.

Credentials already in the wild. Infostealer logs are cheap and searchable. If a staff member's home laptop got popped, your M365 or VPN credentials may be sitting in a dataset right now. AI makes sorting through billions of those records trivial.

Default and shared passwords. Printers, cameras, NAS boxes, building management systems, the admin panel on that one appliance. Still everywhere.

Assets you forgot you have. The test server from the 2022 project. The marketing microsite on an old CMS. A subdomain pointed at something that doesn't exist anymore. You can't patch what you don't know about, and attacker recon is now better than most companies' asset inventories. VulnCheck found WordPress and other content management systems account for about a third of known exploited vulnerabilities.

End-of-life systems. No patches are coming. Ever. If it's internet-facing, it's a countdown.

Weak email authentication. Missing or permissive SPF, DKIM and DMARC make spoofing your domain easy. And AI-written phishing doesn't have the typos that used to give it away.

Notice that none of this requires a nation-state.

One honest caveat

The sky isn't falling. VulnCheck's first-half 2026 data still puts the median time from CVE publication to known exploitation around 80 days, and its researchers think AI is currently helping defenders find and fix bugs more than it's helping attackers exploit them.

Fair point. But medians hide the tail, and the tail is where the damage lives. The same report found roughly 1 in 4 exploited vulnerabilities were hit within a day of the CVE going public. If yours is in that quarter, the median doesn't matter.

The defenders who benefit from AI are the ones using it. Everyone else gets the attacker's side of the trade.

Stop chasing the exotic

Most security budgets I see are tilted toward the interesting problems. New tools, new dashboards, a threat intel feed nobody reads. Meanwhile the VPN still doesn't have MFA.

Vendors are part of the reason. Basic hygiene doesn't demo well, and nobody gets a keynote slot for "we turned on MFA everywhere and killed our exposed RDP." So boards get sold the shiny thing, because the shiny thing has a sales team.

Fix the boring stuff first. That's where attackers are looking, because that's where the easy money is.

What to prioritize, in order

1. Find everything you have on the internet. Run an external attack surface scan, compare it to what you think you own, and chase every surprise. Do it continuously, not once a year.

2. Patch or isolate edge devices first. Firewalls, VPNs, remote access gateways and anything else internet-facing go to the front of the line. Criticals get done in hours. More on that in my next piece.

3. Put phishing-resistant MFA on every remote access path. VPN, RDP gateways, email, admin portals, vendor access. No exceptions for executives.

4. Kill exposed RDP and management interfaces. Nothing administrative should be reachable from the open internet.

5. Watch for your credentials in infostealer and breach data. When they show up, reset and revoke sessions the same day.

6. Hunt down default and shared passwords. Start with network gear, cameras, printers and building systems.

7. Retire or wall off end-of-life systems. If it can't be patched, it can't be on the internet.

8. Lock down your email domain. Enforce SPF, DKIM and DMARC at p=reject.

9. Review third-party access. Every vendor connection into your environment needs an owner, MFA and a reason to exist.

10. Make sure you'd see it. EDR on endpoints and servers, logging on edge devices, and someone actually watching. Assume something gets through.

11. Test it the way an attacker would. Use automated, AI-assisted testing against your own perimeter so you find the easy stuff before someone else does.

None of this is glamorous. All of it works.

AI didn't invent a new way in. It just made the old doors easier to find. Close them.

Next Step

Ready to strengthen your organization's resilience?

A 30-minute discovery call to discuss your cybersecurity posture, incident readiness, and whether advisory support is the right fit.