You’ve probably heard “AI is changing hacking” enough times that it’s started to sound like background noise, another scary headline that may or may not apply to a business your size. Here’s the problem with tuning it out: the mechanics actually did change, and the advice you learned to protect yourself with is built for attacks that don’t work the same way anymore.
How Hackers Used to Attack Businesses
The Old Playbook of Generic Phishing and Cold Calls
For years, cybercriminals ran a numbers game. They blasted the same generic message to thousands of inboxes at once, hoping a small percentage would click. The message usually came from “your bank” or “IT support,” written in stiff, slightly off English, often with a typo or two baked in. A cold call followed a similar script: a stranger reading from a template, fishing for whoever happened to answer and happened to be gullible that day.
It worked often enough to be profitable, but it had a built-in weakness. The attacker couldn’t tailor the pitch. Everyone got the same net, and most people were too specific a fish to get caught in it.
Why the Old Warning Signs Used to Work
That’s where the classic advice came from:
- Watch for bad grammar
- Watch for a generic greeting like “Dear Customer”
- Watch for an email address that almost matches your bank’s but not quite
These tells worked because they traced back to one root cause. A scammer writing at volume, in a hurry, often in a language they didn’t speak fluently, had no reason to polish any single message.
Take away that root cause, and the tells disappear with it. That’s exactly what happened.
Real AI Attacks Are Already Happening
Documented Incidents, Not Hypotheticals
This isn’t a future-tense warning. Anthropic’s own threat intelligence team documented a 2025 case where an attacker used an AI coding assistant to run reconnaissance, write and disguise malware, and draft extortion notes, largely on its own, against 17 different organizations. The human behind it didn’t need deep technical skill. The AI supplied it.
That single case tells you what changed:
- Reconnaissance on target organizations, handled by the AI tool itself
- Malware creation and obfuscation, written and rewritten by the AI to slip past detection
- Extortion notes, drafted and personalized without a human writer
The Gap Between the Old Playbook and Today’s Attacks
The old playbook needed a skilled writer, a convincing voice actor, or a patient con artist willing to build trust over weeks. AI collapses all three into a tool anyone can rent access to. The attacks look the same on the surface (an email, a call, a text), but the labor behind them has changed completely, and that’s the part small business owners haven’t caught up to yet.
AI Has Made Phishing Emails Nearly Impossible to Spot
The End of the Bad-Grammar Tell
A Harvard-affiliated study found AI-automated phishing achieved a 54% click-through rate, more than four times the 12% rate of generic phishing, and on par with attacks written by trained human social engineers. The grammar-check trick your team learned five years ago doesn’t just work less well now. It doesn’t work at all. That shift is exactly what we cover in our deep dive on why phishing emails get past your filters in 2026.
How AI Personalizes an Attack Using Public Information
Modern phishing pulls from what’s already public. AI stitches those scattered details into one message that references your actual world, not a generic template:
- Your job title and recent posts on LinkedIn
- A press release or product launch your company announced
- A vendor or software tool you’ve mentioned publicly
It’s not guessing that you might use a particular vendor. It reads that you do, and it writes the email accordingly.
Deepfake Voices and Video Are Fueling Executive Impersonation Scams
What an AI-Cloned CEO Call Sounds Like
Picture a call from your CFO’s actual voice, asking for an urgent wire transfer before a deadline. That’s not a hypothetical. The FBI’s IC3 2025 Annual Report logged more than 22,000 AI-related complaints and $893 million in AI-linked losses last year, and flagged voice-clone “distress scams” alone at more than $5 million in reported losses. A few seconds of audio, pulled from a podcast appearance or a webinar recording, is often all it takes to build a convincing clone. Our guide on stopping AI CEO impersonation scams walks through exactly how these calls get built.
Why a Video Call Isn’t Safe Verification Anymore
“Let’s hop on a video call to confirm” used to be the gold standard for catching a scam. It isn’t anymore. Real-time deepfake video is good enough to fool a quick glance, especially over a laggy connection where small visual glitches are easy to explain away. Verification now has to happen through a second channel entirely, not a nicer-looking version of the same one.
AI Is Scaling Social Engineering Beyond What One Attacker Could Do Alone
Real-Time AI Chatbots Are Running the Con
A single scammer used to be limited by how many conversations they could juggle in a day. AI chatbots remove that ceiling. One operator can now run dozens of live, personalized conversations simultaneously, each one adapting in real time to what the target says back.
How Attackers Build Trust Faster With AI
The numbers back up what that shift looks like in practice:
- 62% of confirmed breaches involve a human element, per the Verizon 2026 Data Breach Investigations Report, which points to AI as an explicit factor in social engineering, malware development, and target selection
- 4.5x more likely to be clicked: Microsoft’s 2025 Digital Defense Report found AI-assisted phishing gets clicked at nearly five times the rate of the manual version
- Up to 50x more profitable at scale, per the same Microsoft research
Trust that used to take a scammer weeks to build now takes an AI model a few exchanges.
Hackers Are Using AI to Write and Adapt Malicious Code
What AI-Generated Malware Looks Like in Practice
Beyond writing convincing messages, AI is now writing and adapting the malicious code itself. That’s exactly what happened in the Anthropic-documented case referenced earlier: an AI tool that could write malware, then rewrite it on the fly to slip past detection, without a human coder guiding each step.
How AI Lowers the Skill Bar for Attackers
The result isn’t just more attacks. It’s more kinds of attackers:
- Someone with no coding background can now direct an AI tool to produce functional malicious code
- The same way someone with no design background can now ask an AI tool for a logo
- The barrier that used to keep casual criminals out of serious cybercrime is a lot lower than it used to be
How to Protect Your Business From These AI Attacks
The instinct to look for a new set of stylistic tells is understandable, but it’s chasing the wrong target. AI-generated scam content today reads clean, and there’s no reliable, documented pattern that flags it on sight. The defense that holds up isn’t a smarter way to read the message. It’s a smarter way to verify before you act on it.
- Train your team to expect polished, personalized attacks, not obviously fake ones
- Require a second-channel verification step for any request involving money, credentials, or sensitive data, even when the voice or the video looks right
- Layer multi-factor authentication and active monitoring on top of training, since no single defense catches everything alone
Combine Training, MFA, and Monitoring Into One System
None of these tools work well in isolation. Training without monitoring means you’re relying entirely on your team catching every attempt, every time. Monitoring without training means your systems flag the problem after someone already clicked. The businesses holding up best against AI-driven attacks treat training, authentication, and monitoring as one connected system, not three separate checkboxes.
Build a Verification Habit That Sticks
The single habit worth building into your team’s muscle memory: if a request involves money, credentials, or sensitive data, verify it through a channel the attacker doesn’t control. That could be a callback to a known number, an in-person check with someone else on the team, or a separate thread you already trust. It’s a small habit, and it stops nearly every version of this attack cold, whether it arrives as an email, a call, or a video.
Understanding the Threat Is the First Step Toward Stopping It
None of this means the old advice was wrong. It means the ground shifted, and the businesses keeping up are the ones who shifted with it. AI hasn’t changed what hackers are after. It’s changed how convincingly they can ask for it.
If this is something you’re still working through, our team is glad to talk it over.
