Skip to content

Incident Breakdowns

The 72-Hour Reset: 5 Practical Rules When Attackers Move Faster Than You

Pre-disclosure exploits, AI-driven intrusions and leaked developer secrets, all in one week. Five rules for resetting your security baseline, and how each one shows up on a cyber insurance application.

Over the past two weeks, the threat landscape sent an unmistakable signal: the gap between discovery, weaponization and compromise has collapsed to almost nothing.

Consider what collided in a single week:

  • Pre-disclosure exploitation. Amazon caught an attacker trying PHP2Shell four full days before PHP could publish the fix. (Our breakdown of PHP2Shell.)
  • Machine-speed campaigns. A China-state aligned group used an agentic AI framework and stolen ConnectWise app credentials to attempt sign-ins to more than 2,250 Microsoft 365 tenants, in some cases reaching privileged access in under three hours. (Our breakdown of Storm-0940.)
  • Developer secrets targeted. A platform breach at Railway put environment variables in up to 100,000 workspaces at risk. (Our breakdown of the Railway breach.)

If your team is operating on the assumption that you have 30 days to patch critical bugs, or that humans can manually triage active identity intrusions, you’re using a playbook from 2018.

Here is how to reset your baseline for the reality of late 2026.

1. Fix the timelines in your incident playbooks

When a high-profile CVE drops, the natural instinct is to search your logs starting from the announcement date. Stop doing that.

Amazon’s honeypots caught a PHP2Shell attempt on October 17. The public didn’t know until October 21. When hunting for exposure, pull your timeline back at least five to seven days before disclosure. If you only look at traffic from the day of the news, you’re ignoring the first breach window.

2. Treat third-party app grants like root access

You probably have tight controls around who gets domain admin or root credentials. But who watches the OAuth apps connected to your Microsoft 365 or Google Workspace tenant?

The ConnectWise incident proved attackers don’t need to crack your CEO’s password if they can ride an enterprise app identity straight through Entra ID.

  • Audit your SaaS integrations quarterly.
  • Remove apps that haven’t been used in 90 days.
  • Use conditional access rules that restrict app identities to expected locations and IP ranges.

3. Kill the “we’ll patch next weekend” mindset

When the FBI, the NSA and CISA all tell organizations to take unpatched PHP servers offline, that’s a warning about systemic risk. If you have legacy systems on end-of-life versions like PHP 7.4 or 8.1, there’s no official release to wait for. You either isolate the machine, put a strict reverse proxy or WAF in front of it while you upgrade, or accept that the box will eventually belong to someone else.

4. Assume your long-lived API keys are leaked

If your developers paste raw API tokens and database URIs into plain-text configuration panels, expect them to leak eventually.

  • Require IP restrictions on sensitive keys, such as payment gateways and database servers.
  • Set automatic expiration dates on service credentials.
  • If a vendor reports anomalous activity, rotate your secrets before waiting for its final forensics report.

5. Build for machine-speed defense

Humans are great at strategy, root-cause analysis and communication. We are terrible at stopping an automated agent working through hundreds of sign-ins at 2:00 a.m. on a Sunday.

Your defenses need automated kill switches. If an identity suddenly starts making hundreds of Graph API calls or pulling mailbox data from unusual endpoints, your identity provider should revoke that session automatically, not email an analyst to ask whether it looks normal.

Where cyber insurance fits

Every one of these rules shows up on a cyber insurance application in some form: how fast you patch critical vulnerabilities, whether you run end-of-life software, how you govern third-party access, whether you use MFA and endpoint detection, and whether you have a written incident response plan. The answers shape your price, and in some policies they shape what gets paid.

A good policy is also part of the response plan itself. Most require prompt notice and give you a breach coach, forensics and legal counsel within hours. Knowing who you call, and in what order, is worth sorting out on a quiet Tuesday rather than a loud Sunday.

The bottom line

Security in 2026 isn’t about building an impenetrable fortress. It’s about shrinking your blast radius and closing the gap between detection and containment. Keep your architecture simple, know where your secrets live, and never let legacy plumbing sit forgotten in the corner.

This article is educational and is based on public reporting at the time of writing. It does not alter any policy. Coverage is determined by the actual policy forms, endorsements and declarations issued by the insurer.

I Don’t Need Cyber Insurance

More from CIG Cyber Chronicles

All articles

Think you don’t need cyber insurance? Let’s test that against how your business actually runs.

Schedule a Coverage Review