Here's a number that should make you pause: SentinelOne reports a 1,265% increase in AI-generated phishing attacks over the past year. That's not a typo. Thirteen-fold. And unlike the clumsy "Nigerian prince" emails of years past, today's AI lures are personalised, grammatically flawless, and frighteningly convincing.

We've been tracking this surge across credential databases, threat reports, and incident response data. The picture is clear: 2026 is the year AI turned phishing into an industrial-scale operation.

Why AI Phishing Is Different This Time

Phishing isn't new, but the tools have changed. Generative AI now lets attackers craft emails, texts, and even voice calls that mimic real people with unsettling accuracy. Hoxhunt's 2026 phishing trends report documents a 14x end-of-year surge in AI-generated attacks, with new vectors like SVG file attachments and malicious calendar invites.

The economics explain why. A phishing campaign that used to take hours of manual effort now takes seconds with a language model. Attackers feed an LLM a target's LinkedIn profile, a recent company blog post, or a data-breach dump, and the model produces a pitch-perfect impersonation email — complete with the right tone, industry jargon, and personal details.

How Attackers Get the Raw Material

Infostealer malware is the engine behind this. KELA's State of Cybercrime 2026 report found that 2.86 billion credentials were compromised in 2025 alone, including passwords and session cookies. These credential dumps — often from stealer logs — give attackers everything they need to personalise phishing at scale.

Once an attacker has your email from a breach like the January 2026 infostealer database (149 million credentials, including 48 million Gmail accounts), they can cross-reference it with public profiles and send you a phishing email that references your real accounts, your real employer, and your real recent activity.

In our testing, we fed a publicly available breach sample into an uncensored language model and asked it to write a phishing email targeting a specific company. The result was indistinguishable from an internal IT notice — correct formatting, correct terminology, no spelling errors. That capability is now available to anyone with an internet connection and US$200 for a malware kit.

Canvas Breach: The Perfect Phishing Trigger

The ongoing Canvas security incident — where ShinyHunters stole 3.65 TB of data from Instructure affecting 8,809 institutions — shows exactly how this plays out. 231 million unique email addresses belonging to students, teachers, and staff are now circulating among threat actors. The deadline for ShinyHunters to leak the data is May 12, 2026, but the emails are already out there.

We expect a wave of targeted phishing in the days ahead: fake Canvas login pages, fake IT help-desk messages, fake assignment links. The FBI's IC3 has historically warned of phishing spikes within 72 hours of major breaches, and this one is the largest educational breach on record.

What AI Phishing Looks Like in 2026

Based on data from Huntress, Hoxhunt, and our own monitoring, here are the most common AI-powered phishing formats right now:

Deepfake Voice Calls (Vishing)

Attackers clone a voice from a 30-second social media clip and call targets posing as a colleague, IT support, or bank representative. The Federal Trade Commission reported a sharp rise in voice-cloning scams in 2025, and the trend has accelerated in 2026. A 2026 Forbes investigation found that AI-generated voice scams now account for 1 in 3 reported vishing incidents.

Personalised Spear-Phishing Emails

Generic "Dear Customer" emails are disappearing. AI generates emails addressed to you by name, referencing your specific accounts, even mentioning your manager by name. These emails pass through traditional spam filters because they contain no malware and no suspicious links at first — the payload comes in a follow-up message after the target responds.

SMS Phishing (Smishing) with Shortened Links

AI drafts SMS messages that sound like real notifications from your bank, delivery service, or school. Shortened URLs (bit.ly, tinyurl) bypass link scanners. CyberProof's 2026 report warns that identity-based attacks — where stolen credentials are used to authenticate into real services — are the fastest-growing category.

How to Spot AI-Generated Phishing

AI phishing is harder to spot than traditional phishing, but not impossible. Here's what we look for:

Urgency Without Context

AI models love creating urgency because it works. "Your account will be suspended in 24 hours" or "Unauthorised login detected — verify now." Legitimate organisations give you time. If an email demands action within hours, it's a red flag.

Requests for Credentials

No legitimate service asks for your password by email, text, or phone. Not your bank. Not your IT department. Not Canvas. If someone is asking for your password, they are trying to steal it.

Anomalous Sender Addresses

AI can write perfect copy but can't always spoof the domain. Check the sender address carefully. "[email protected]" is not the same as "[email protected]." Hover over links before clicking — the display text and the actual URL should match.

Unsolicited Attachments or Calendar Invites

Hoxhunt reports a sharp rise in SVG file attachments and malicious calendar invites sent via phishing campaigns. If you weren't expecting a file or meeting invite, don't open it.

The Anti-Phishing Toolkit: 5 Layers of Protection

Based on NIST SP 800-63B guidelines and CISA recommendations, here's the layered defence every person should have in 2026:

1. Use a Password Manager

Password managers like Bitwarden, 1Password, and Keeper don't just store passwords — they autofill them only on the correct site. If you land on a fake Canvas login page, your password manager won't offer to fill in your credentials because the URL doesn't match. That alone stops most credential phishing dead.

Try our free password generator to create unique, strong passwords for every service — then store them in a password manager.

2. Enable Phishing-Resistant MFA

Standard MFA (SMS codes) blocks 99% of automated attacks, but AI-powered real-time phishing can bypass it. FIDO2 hardware security keys (YubiKey, Google Titan) are phishing-resistant because they verify the website's identity before authenticating. CISA and the UK's NCSC both recommend FIDO2 as the gold standard in 2026.

3. Use a Breach Monitoring Service

Check your email on Have I Been Pwned to see if it appeared in recent breaches. The January 2026 infostealer database and the Canvas breach both add millions of new records to the list. If your credentials are in a known breach, change those passwords immediately.

4. Run Antivirus with Real-Time Protection

Infostealer malware — the kind that feeds AI phishing pipelines — needs to be on your device first. A good endpoint protection tool (Bitdefender, Malwarebytes, Windows Defender) catches most stealer variants before they can exfiltrate your saved passwords and session cookies.

5. Verify Out-of-Band for Sensitive Requests

If you receive an email from your "IT department" asking you to log in to a portal, pick up the phone and call them. If your "bank" calls asking you to verify a transaction, hang up and call the number on the back of your card. Any request involving credentials, money, or sensitive data should be verified through a separate channel.

What Organisations Should Do

The CyberProof 2026 report warns that identity-based AI attacks are outpacing most organisations' defences. For IT and security teams managing password policies, the NIST SP 800-63B guidelines recommend:

• Mandatory compromised-credential screening — reject any password found in known breach databases
• Phishing-resistant MFA for all privileged accounts
• Security awareness training that covers AI-generated phishing specifically
• Automated phishing simulation with AI-generated templates (not the old "click here for a prize" variety)

For IT administrators managing password policies across their teams, our blog covers NIST and NCSC-aligned password policy guidance that can generate compliant password rules.

FAQs About AI Phishing in 2026

What is AI phishing?

AI phishing uses generative AI — large language models, voice cloning, and deepfake technology — to create convincing phishing emails, text messages, and phone calls. Unlike traditional phishing, AI-generated attacks are personalised, grammatically correct, and often indistinguishable from legitimate communications.

How much has AI phishing increased in 2026?

SentinelOne reports a 1,265% increase in AI-generated phishing attacks over the past year. Hoxhunt documents a 14x surge in AI phishing volume by end of 2025, and the trend has accelerated further in 2026.

Can AI phishing bypass two-factor authentication?

Yes. Real-time AI phishing tools can intercept MFA codes by acting as a proxy between the user and the legitimate site — the user logs in, enters their code, and the attacker uses it immediately. This is called an "adversary-in-the-middle" attack. FIDO2 hardware security keys are resistant to this type of attack.

How do I know if I've been phished by AI?

Check your login history on critical accounts (email, banking, school portals). Look for sign-ins from unfamiliar locations, devices, or IP addresses. Run a malware scan to check for infostealer infections. If you entered your credentials on a suspicious page, change them immediately and enable MFA.

Does the Canvas breach mean I'll get phishing emails?

It's highly likely. The 231 million unique email addresses stolen in the Canvas breach are already circulating. Expect fake Canvas login pages, fake IT support messages, and fake assignment links in the coming days. Never click a link in an unexpected email — navigate directly to canvas.instructure.com instead.

What's the single best defence against AI phishing?

A password manager. It won't autofill your credentials on a fake website, even if the page looks identical to the real one. Combined with a FIDO2 security key for your most important accounts, a password manager is the closest thing to a silver bullet against credential phishing.

Are traditional spam filters effective against AI phishing?

Partially, but not reliably. AI-generated emails don't contain the spelling errors, odd grammar, or suspicious formatting that traditional filters look for. Advanced filters using natural language processing perform better, but no filter catches everything. Human vigilance remains essential.

How do attackers get my email for AI-phishing campaigns?

Data breaches are the primary source. The January 2026 infostealer database contained 149 million credentials, the Canvas breach leaked 231 million emails, and KELA reports 2.86 billion total credentials compromised in 2025. Any of these dumps likely contain your email address. Attackers also scrape public LinkedIn, company websites, and social media profiles for targeting data.

To defend against AI-powered phishing attacks, a comprehensive security suite like Kaspersky Premium provides real-time phishing detection, malicious link blocking, and credential theft protection across your devices.