Log inGet a Demo

Eight Days: What the BSI's 2026 Report Means for Your Patch Cycle

The BSI raised its rating of IT security in Germany to "tense to critical." The more important change is in the numbers: attackers now exploit newly published vulnerabilities about 8 days after disclosure, down from 53 a year ago. Here is what that means for a monthly patch cycle, and what to do now.

Patrick Münch
Patrick Münch
·6 min read·
Eight Days: What the BSI's 2026 Report Means for Your Patch Cycle

The BSI published its annual report on the state of IT security in Germany this morning, and the rating went up from "tense" to "tense to critical." That's the line in every headline today.

It's the right call. But the BSI has used those words before, in 2021 and in 2023. If you only read the rating, you'll miss what's actually new this year. For me it's one number on page 15.

What the BSI found

A few figures from the report (it's in German for now):

  • In Q2 2026, attackers exploited newly published vulnerabilities about 8 days after disclosure on average. For the full reporting period it was 25 days. In last year's report it was 53.
  • About 160 new vulnerabilities were published per day, up 35%.
  • Exploitation attacks rose 48%. Scans for vulnerable systems on the internet rose 57%. New malware sent by email went down about 20%.
  • The vulnerable attack surface the BSI measures grew 117%, mostly because of IoT devices and AI.
  • 90% of companies surveyed rate their own security "good" or "very good." Companies that went through a BSI check met only 55% of the minimum requirements on average.

Put together, attackers have moved away from tricking people into clicking on something. They go straight for whatever is reachable from the internet, and they get there faster every quarter.

Eight days against a monthly patch cycle

Here's why the eight days matters so much. Most organizations I talk to still patch on a monthly rhythm. A fix comes out, it gets triaged, it waits for a change window, somebody signs off, and it goes in. Our State of Vulnerabilities research puts the median time to patch at about 32 days.

So on a typical critical bug, the attacker is ready in about a week, and you finish about three weeks after that. For most of the month you're exposed and the clock is running.

And eight days is an average. The BSI says that for a lot of vulnerabilities the time is now down to a few days or even hours.

Why AI changed the math for attackers

Writing a reliable exploit used to be slow, skilled work. A specialist would spend days or weeks on it. The BSI says that's gone. Attackers now have exploits ready almost as soon as a vulnerability has been analyzed, and the analysis itself is automated.

There are a few reasons. Agentic AI can find a vulnerability and build the exploit for it at the same time, and it can chain several weaknesses together into a new way in. Attackers also reverse-engineer patches. They compare the fixed version with the old one and work back to the bug, so the patch release itself becomes the starting gun. The report cites research where a large language model, pointed at a widely used web browser, went from finding about one high-severity vulnerability a week to about eight within a few months, and could write the exploit too.

The report includes two details I keep coming back to. In autumn 2025 the BSI saw an automated, AI-driven attack. And this summer, AI agents broke out of their test environments and attacked third parties who had nothing to do with the test.

BSI President Claudia Plattner said today she expects the next stage to bring more attacks by AI agents acting on their own. The report also expects that as early as next year most exploited vulnerabilities will be zero-days, with no patch available when the attack starts.

The confidence gap

The 90% and 55% figures bother me more than any of the attack numbers.

Nine out of ten companies think they're in good shape. The companies the BSI actually assessed, which are the ones paying enough attention to ask for a check, met only about half of the basics. Most of the gap is ordinary stuff. Systems nobody knew were exposed, patches that never got scheduled, old services still running because nobody owns them.

An attacker running an automated scan doesn't care how confident you are. They see an exposed service with a known vulnerability, and they take it.

Compliant isn't the same as fast

A lot of German organizations spent this year on NIS2. They registered with the BSI, wrote risk management policies and set up incident reporting. That work matters, and the BSI says NIS2 has come at the right time.

But an audit asks whether you have a patch process. It doesn't ask whether that process can close a critical bug on an internet-facing system within a week. With exploitation at eight days, those are two different questions, and a lot of companies can say yes to the first and no to the second.

What I'd do now

Know what faces the internet. You can't patch fast what you don't know you have. Build the list of everything reachable from outside, and include the things nobody thinks of as servers: appliances, VPN gateways, test systems, forgotten cloud instances, and the AI tools teams set up on their own. Then cut it down. The BSI's advice is to keep the attack surface as small as possible and limit access to trusted sources wherever a service doesn't need to be open to everyone.

Set a separate clock for exposed systems. A monthly window is fine for a lot of your estate. It isn't fine for a critical vulnerability on something an attacker can reach. The BSI says patches for critical and high vulnerabilities on exposed systems shouldn't be deferred. Give yourself a target in days and measure against it.

Automate the rollout. The BSI recommends automated patching, using AI where it helps. In my experience the time doesn't go into finding the problem. It goes into everything after that: working out who owns the system, testing the fix, waiting for approval, rolling it out and checking it worked. That's where automation pays off.

Start with what everything else depends on. Not every exposed system carries the same risk. Login and single sign-on, remote access, build pipelines and anything holding credentials go first, because an attacker who gets in there doesn't just have one server. Then keep going through the list.

If you patched late, check for compromise. The BSI says this plainly. If a system was exposed for weeks before the fix went in, patching it doesn't undo anything that already happened. Look for signs that someone got in first.

Treat your own AI agents like any other system. If an agent has credentials and permission to run things, it's part of your attack surface. Know where it runs, keep its permissions tight and patch the software underneath it.

One question for your next team meeting

If you take one thing into your next team meeting, make it this: when a critical vulnerability is published on a Tuesday, on what day is it actually fixed on our internet-facing systems?

If the answer is "the next maintenance window," you now know where to start.

About the Author

Patrick Münch

Patrick Münch

Co-Founder & CSO

Chief Security Officer (CSO) at Mondoo, Patrick is highly skilled at protecting and hacking every system he gets his hands on. He built a successful penetration testing and incident response team at SVA GmbH, their goal to increase the security level of companies and limit the impact of ransomware attacks. Now, as part of the Mondoo team, Patrick can help protect far more organizations from cybersecurity threats.

Ready to Get Started?

See how Mondoo can help secure your infrastructure.