AI is about to make cyberattacks faster, more convincing and easier to launch, and the window to get ready is short. This year, AI systems proved they can find serious software flaws on their own. Defenders got that capability first. Attackers won't be far behind.
In August, MSP Today put it bluntly: businesses may have about six months to prepare. Having looked at the research behind that warning, we agree. Here's what's changing, and three practical priorities to work on between now and spring.
AI can now find the holes on its own
In April 2026, Anthropic announced Claude Mythos Preview, a model it says found thousands of high-severity zero-day vulnerabilities across every major operating system and web browser. A zero-day is a flaw nobody knew about, so no patch exists yet.
Some of the bugs had hidden in plain sight for years:
- A 27-year-old flaw in OpenBSD that let an attacker remotely crash any machine running it.
- A 16-year-old bug in FFmpeg, video software used almost everywhere, that automated testing had run past five million times without catching.
- Several Linux kernel flaws chained together to take full control of a machine.
Anthropic chose not to release the model publicly. Instead, through Project Glasswing, it gave access to Microsoft, Google, Apple, the Linux Foundation and others so they could fix flaws first.
That head start won't last forever. Cloud security firm Wiz estimated 12 to 18 months before comparable ability shows up in open models that anyone can download and run.
Attackers were already fast
Even before AI, attackers were exploiting flaws faster than most businesses patch them. Mandiant's M-Trends 2026 report shows how fast:
−7 days: the average time to exploit
Flaws are typically exploited a week before a patch or public disclosure even exists.
32% of break-ins start with an exploited vulnerability
It's the single most common way in.
17% of break-ins start with voice phishing
A phone call is the second most common way in.
22 seconds to hand off your network
When one criminal group passes access to your network to another, like a ransomware crew, the handoff used to take more than 8 hours (2022). By 2025 it took 22 seconds.
Now add AI that can find and weaponize flaws in hours.
What AI changes for attackers
AI gives attackers three advantages.
- It finds flaws faster. Research that once took an expert weeks can now take hours, working exploit included.
- It fools people better. Phishing emails arrive with perfect grammar and real project names. Voices can be cloned, so a call from "the CEO" or "IT support" can sound exactly right.
- It lowers the skill bar. AI models with the safety guardrails removed have been downloaded millions of times. Running a sophisticated campaign no longer takes an expert.
The blind spot inside: shadow AI
Shadow AI means AI tools your staff use without IT knowing, and it is already causing breaches. IBM's Cost of a Data Breach Report 2025 found that one in five organizations had a breach linked to shadow AI. Those breaches cost an average of US$670,000 more, and only 37% of organizations have policies to manage or detect it.
In a typical office, shadow AI looks like this:
- Client or financial data pasted into a free chatbot
- AI browser extensions that can read every page an employee opens
- AI note-takers joining Teams calls and storing recordings outside your control
- AI agents connected to Microsoft 365 with broad access to mail and files
- Old accounts and app connections nobody remembers setting up
Nobody is doing this to cause harm. People are trying to work faster. But every unmanaged tool is a door you can't see.
Three priorities for the next six months
The response comes down to three priorities: patch faster, govern shadow AI, and verify people instead of trusting voices.
1. Patch faster than AI can exploit
A monthly patch cycle is no longer enough for everything.
- Shrink what's exposed. List everything reachable from the internet, and remove whatever doesn't need to be there.
- Create an emergency patch lane. Fix critical flaws on internet-facing systems within 48 hours, using maintenance windows approved in advance.
- Plan for when there's no patch. Disable the affected feature, block access, or isolate the system until a fix arrives.
- Retire end-of-life software. Anything that no longer receives updates is a permanent open door.
2. Bring shadow AI into the light
The goal isn't to ban AI. It's to make sure people use it safely.
- Discover: find out which AI apps, browser extensions, note-takers and Microsoft 365 connections are actually in use.
- Decide: agree on approved tools, such as Microsoft Copilot or business versions that don't train on your data.
- Document: write a short, plain-language AI policy covering what data can and can't go into AI tools.
- Guard: add sensitivity labels and data loss prevention rules, block unapproved uploads, and remove stale accounts.
3. Verify people, not voices
Voice phishing is already the second most common way in, and AI voice cloning makes it far more convincing.
- Callback rules: verify any request to change payment details, send money, or reset a password or MFA by calling a number you already have.
- Code words: give executives and finance staff a verbal code for urgent requests.
- Phishing-resistant MFA: move admins, finance and executives to passkeys or hardware security keys.
- AI-era training: show staff real examples of AI-written phishing and cloned voices before they meet the real thing.
A simple six-month plan
Split into three phases, the work stays manageable alongside everything else.
- November to December: assess. Scan what's exposed to the internet, take stock of the AI tools in use, and review how quickly patches go out and how payment requests are verified.
- January to February: fix. Set up the emergency patch lane, publish an AI policy and approved tools, roll out passkeys for high-risk users, and clean up stale accounts.
- March to April: test. Run a tabletop exercise built around an AI scenario, such as a cloned-voice CEO calling finance. Run a phishing test with AI-written emails, then re-check where you stand.
By May 2027, the goal is to be ahead of AI-powered attacks rather than reacting to them.
Use the head start
Defenders got a head start this year, and the next six months are about using it. None of these steps needs a big budget. Each one closes a door that AI is about to make easier to open.
Want help getting ahead of AI-powered attacks?
Talk to the HyCloud team. We’re happy to show you where your business stands.
