What a time to be an MSP.
N-central got hit with a pile of zero days, pushed hotfixes, and the next night another batch dropped. Alarm bells rang as ScreenConnect was used in a series of attacks. MSP Process got breached, and attackers walked off with the keys to its customers' Microsoft 365 tenants. A couple of weeks before all that, a stolen ConnectWise Cloud Backup credential (the old SkyKick product) got attackers into 173 Microsoft 365 tenants in about two hours. No password, no MFA.
One of those happening is a bad week. All of them stacked on top of each other is a pattern.
And that's before the Microsoft outage, the FAA ground stop, and today's bank outages. As far as we know, none of those are attacks. But when everything's breaking at once, it gets hard to tell the difference, and your clients are going to ask you which one it is.
Why now?
The N-central bugs probably weren't even in new code. They could have been sitting there for years. Anybody could have found them at any point. Nobody did, until now. And this is not a problem unique to N-central.
So what changed? My bet is AI. I don't know that anyone used AI to find the N-central bugs. But the tooling is out there, it's cheap, and it's really good at chewing through old code nobody has looked at in years. Heck, some of it has been around so long people probably thought it WAS secure. When finding a bug gets that cheap, the backlog doesn't stay buried.
The numbers back that up. CVE disclosures passed 100,000 this year, more than double all of 2025. For most of the last decade the count grew about 16% a year. Then Anthropic announced Mythos in April, and monthly CVE volume roughly tripled by October. VulnCheck, which tracks this closely, puts the surge on AI-assisted bug hunting, and expects 100,000 to look like a quiet year in hindsight.
Not every one of those is a five-alarm fire. The share actually getting exploited has held steady at around 1.4%. But 1.4% of a number that doubled is still a lot more to patch.
I think we're at the very beginning of AI working through a decade's worth of technical debt in the platforms we use to run our clients. And we're going to keep getting hit like this, over and over, for the next several months.
The ConnectWise and MSP Process hits are a different flavor of the same problem. Those weren't zero days. They were stolen app credentials with standing access to client tenants, a lot of it granted years ago and never cleaned up. That's technical debt too, it just lives in the tenant instead of the code.
Where do we go from here?
Right now we're behind the curve. Attackers move a lot faster than defenders, and it's time to get off our heels. Here's what I think happens next.
Bug bounty programs should ramp up. The tools are available to anyone now, so more researchers come into the market. Most vendors can't staff up fast enough to do this work internally, so paying outsiders becomes the only realistic way to keep up. And for a lot of the people finding these bugs, this isn't about morality, it's about money, and right now crime pays better. A bounty program that pays like it means it won't turn organized crime into good guys. It can flip the next person deciding which side to sell to.
Patch windows collapse. Patch timing has always been a risk calculation. We measured the likelihood of a bad patch against the likelihood that a bad guy found the thing that was unpatched. That calculation just drastically changed. A monthly patch cycle made sense when a bug took weeks to weaponize. Exploitation now starts sooner after disclosure every year, and AI that can find a bug can help write the exploit too. A 30-day patch window is just a 30-day exposure window. If your patch policy still says "next maintenance window" for critical vulns on internet-facing tools, it's already out of date.
Vendor due diligence gets real. Clients, insurers, and auditors are going to start asking about the tools in your stack, not just the tools in theirs. Do your vendors run a bug bounty? How fast did they patch the last critical? Are they pointing the same AI tools at their own code that attackers are pointing at it? Those become RFP questions, and some vendors won't have good answers. Some boats need to sink. Better to find out which ones before your clients are on board.
It's going to get worse before it gets better (and it will get better). Eventually the vendors catch up and start pointing the same tools at their own code before the attackers get there. Once that happens, things calm down, hopefully.
But I don't think calm means back to "normal." Vendors are shipping AI-written code faster than ever, and new code means new bugs. The flood slows down, but we're not in Kansas anymore, and I don't think we ever get back.
What to do today
None of this is new. It's fundamentals. We've just been skipping them for years because nothing forced the issue. That excuse is gone.
- Shrink your attack surface. Some stuff has to face the internet, that's the nature of the business. If it doesn't, take it off the internet.
- Get rid of platforms you don't need. That tool you're paying for and barely touching is still a door someone can walk through.
- Audit the enterprise apps in every client tenant, and your own. ConnectWise and MSP Process both came in through app consents with standing access. If nobody can tell you why an app still has that access, pull it.
- Limit who can get into the platforms you keep. Least privilege. Actually do it this time.
- Limit where it can be reached from. Internet-facing doesn't have to mean open to the whole internet. IP allowlists, geo-blocking, and a VPN or ZTNA in front of admin consoles all count.
- Limit what the tools themselves can do. If it doesn't need admin rights, don't hand it admin rights because it's easier.
- If you self-host, put a WAF in front of it. A lot of us neglected that for years. We can't anymore.
- Any tool that can make changes in a client environment should be sending logs to a SIEM or something like it. When the next one hits, you need to see what happened so you can respond.
- Use AI. This is a gunfight, and if you're a defender who isn't using AI, you brought a spoon. Just don't hand it admin rights across every client while you're at it.
And finally... dust off YOUR business continuity plan
This isn't hypothetical. Last week N-able took hosted N-central offline and told on-prem customers to shut theirs down for a full day, because there was no patch yet. So if your RMM gets compromised tomorrow and there's no patch, how do you turn it off?
And once it's off, how do you keep delivering what you promised your clients?
Shutting down a core platform belongs in your incident response plan as a written step, not something you figure out at 2 a.m. while clients are calling. And running without it is a business continuity problem. Yours, not just theirs.
Last week wasn't a fluke. Plan like it's going to happen again, because I think it will.
If your RMM went dark tomorrow with no patch in sight, how long could you keep your promises to clients?