2,900 Sign-In Attempts Per Minute. Privileged Access in Under 3 Hours. This Is the New Reality of AI-Powered Attacks.

When we talk about AI in cybersecurity, most conversations focus on the defense side - AI-powered threat detection, AI-assisted vulnerability scanning, AI helping overwhelmed security teams keep up with alerts.

That conversation needs to expand. Because the attackers figured out AI first.

Last week, Microsoft disclosed that a China-linked threat group it tracks as Storm-0940 used an open-source AI hacking framework called ARTEX to hit more than 2,250 Microsoft 365 customer environments. At the peak, the AI agents were making 2,900 sign-in attempts per minute. In several confirmed cases, the attackers progressed from initial targeting to privileged cloud access in under three hours.

Three hours. That's the timeline your security team is now operating against.

What Actually Happened

The attack started with a supply chain compromise. Storm-0940 gained access to application credentials from ConnectWise's Cloud Backup service - the same kind of trusted, OAuth-connected vendor application that thousands of organizations grant access to their Microsoft 365 environments every day. Once they had those credentials, they didn't need to guess passwords. They didn't need to trick an employee into clicking a phishing link. They were already inside - because a trusted vendor's identity had been compromised.

From there, ARTEX took over. The AI framework - originally built as an open-source penetration testing tool - automated everything: reconnaissance, scanning, password spraying against additional accounts, OAuth token abuse, and privilege escalation. It conducted over 151,000 failed login attempts across 327 accounts in just four days. And then, using the ConnectWise application credentials, it moved against 2,250 Microsoft 365 tenants at machine speed.

In the environments where it succeeded, AI agents went further - adding backdoor credentials to applications, assigning themselves Global Administrator roles, and modifying group memberships and identities. The changes were designed to persist even after the original ConnectWise application was disconnected. Cleanup alone would require hunting through Entra ID, Purview audit logs, and Microsoft Graph activity going back weeks.

Why This Is Different

Here's the thing about a credential stuffing attack or a phishing campaign: your security team has some chance of catching it before significant damage is done. The attacker is operating at human speed. They're checking results, pivoting, making decisions. There's friction.

ARTEX eliminates the friction. An AI agent doesn't sleep, doesn't lose focus, doesn't make mistakes when it's tired at 2 AM. It executes the attack playbook faster and more consistently than any human team.

"2,900 sign-in attempts per minute and gaining privileged cloud access in under three hours is not a timeline any human analyst queue can keep up with, no matter how well-staffed the SOC," one security researcher put it bluntly.

This is the asymmetry that's keeping security professionals up at night. AI can be used to accelerate attacks far faster than it can accelerate defense. Patching takes time to test, to validate, to deploy. Attackers can exploit a vulnerability the moment they discover it - or the moment a leaked set of credentials lands in their tool.

And ARTEX isn't a cutting-edge nation-state tool anymore. It was open-source. It was freely available for penetration testing. The developers took it private after it was linked to attacks - but that horse has left the barn. Similar frameworks are proliferating.

What This Means for Your Business

If your organization uses Microsoft 365, and you've connected any third-party vendors to it - backup tools, SaaS management platforms, security monitoring services - you need to understand this attack path. Here's what to examine:

Third-party application permissions. Do you have a complete inventory of every OAuth application connected to your Microsoft 365 tenant? If a vendor is compromised, those application credentials can be used to access your environment without a password, without MFA, without any interactive sign-in. If you don't know what's connected and what permissions each application holds, you have a blind spot that attackers are already looking for.

Privileged role monitoring. The ARTEX agents didn't just authenticate and read data. They assigned themselves Global Administrator roles. That kind of privilege change should generate an immediate alert - but only if you're monitoring for it. Many organizations are not. Review your Entra ID audit logs for unexpected role assignments, and make sure any change to privileged roles triggers real-time notification to your security team.

Your patch and detection timeline. In this incident, some organizations were compromised in under three hours. If a breach started right now, how quickly would your team detect it? How quickly could you revoke application credentials and identify what was changed? If you don't have a clear answer to those questions, that's the gap to close first.

The "Black October" Context

Security researchers are calling October 2026 "Black October" - not as a Halloween reference, but because of the volume and sophistication of cyberattacks this month: the PHP zero-day affecting 70% of the internet, the DHS employee data breach, multiple financial sector attacks, the ConnectWise supply chain compromise. These aren't random events. They're a preview of an environment where AI-powered attack tooling is becoming commoditized.

Storm-0940 used ARTEX against organizations across aerospace, defense, financial services, government, legal, manufacturing, energy, semiconductor, and technology sectors. If you're in any of those industries - or if you serve clients who are - you're a plausible target.

Three Questions to Ask This Week

1. What third-party applications have OAuth access to your Microsoft 365 tenant, and what permissions do they hold? If you can't answer this from memory, your IT team should be able to pull a complete service principal inventory. Do it now, before an incident requires it.

2. Who gets alerted when a new Global Administrator role is assigned in your tenant? If the answer is "nobody in real time" or "we'd find out in the next audit," you have a detection gap that this specific attack technique would exploit.

3. What's your actual containment timeline? Not the theoretical one in your incident response plan - the real one, based on how your team actually operates. Machine-speed attacks demand near-machine-speed detection and response. Know your numbers.

The attack tools are getting faster. The only answer is to know your environment better than the attacker does - and to instrument it so that when something changes, you know immediately.

If you're not sure how your organization stacks up against this threat model, that's exactly the conversation TrustPoint Cyber is here to have.