Google's June 2026 Fraud Advisory: What AI-Powered Scams Mean for Your Fraud Stack

On June 19, 2026, Google published its latest fraud advisory. The headline number: a 1,210% increase in AI-powered scam activity over the previous 18 months. The advisory detailed how generative AI tools have fundamentally changed the economics of fraud, making sophisticated attacks accessible to anyone with a laptop and an API key.

This is not a theoretical warning. Google simultaneously announced it had filed a lawsuit against a China-based scam ring that used Gemini AI to generate fraudulent content at scale, including fake investment platforms, phishing campaigns, and synthetic identity documents.

For developers building fraud prevention systems, the advisory is a wake-up call. The attack surface has expanded, the sophistication floor has risen, and traditional defences are failing. Here is what changed, why it matters, and what to do about it.

The Five Threat Vectors Google Highlighted

The advisory breaks down AI-powered fraud into five primary categories. Each one represents a shift from manual, low-volume attacks to automated, high-volume campaigns.

1. Deepfake voice cloning

AI voice cloning tools can now replicate a person's voice from as little as 10 seconds of sample audio. Scammers use cloned voices to impersonate executives in business email compromise (BEC) attacks, call banks to authorize transfers, and manipulate victims in social engineering schemes.

The FBI's Internet Crime Complaint Center (IC3) reported that BEC losses exceeded $2.9 billion in 2025, and early 2026 data suggests voice-clone-assisted BEC is growing at 340% year over year. The barrier to entry has collapsed: open-source voice cloning models are freely available on GitHub, and commercial APIs offer voice cloning as a feature.

2. Synthetic identities

Synthetic identity fraud combines real and fabricated personal information to create identities that pass standard verification checks. An attacker might pair a real Social Security number (often stolen from minors or deceased individuals) with a fabricated name, date of birth, and AI-generated photo.

The Federal Reserve estimated synthetic identity fraud losses at $6 billion annually in their 2025 report, and the trend is accelerating. AI makes it trivial to generate consistent, realistic identity documents, selfie photos that pass liveness checks, and backstory details that withstand manual review.

3. AI-generated phishing

This is the category with the most staggering numbers. According to SlashNext's 2026 Phishing Intelligence Report, 82.6% of all phishing emails are now AI-generated. These are not the crude, typo-filled phishing attempts of the past. AI-generated phishing emails are:

• Grammatically perfect with natural tone and phrasing
• Personalized using data scraped from LinkedIn, company websites, and social media
• Dynamically generated to bypass content-based email filters
• Produced at scale, with thousands of unique variations per campaign

Traditional email security that relies on known signatures or keyword matching is effectively blind to these attacks. The emails look legitimate because they are written by the same language models that power legitimate business communication.

4. Adversary-in-the-middle (AiTM) attacks

AiTM attacks intercept authentication flows in real time. The attacker positions a proxy server between the victim and the legitimate service, capturing credentials and session tokens as they are entered. What makes the 2026 variant dangerous is AI-powered automation: attackers use AI to dynamically replicate login pages, adapt to multi-factor authentication prompts, and complete the entire credential theft process in under 3 seconds.

Microsoft's Digital Defense Report 2026 documented a 146% increase in AiTM attacks targeting cloud authentication flows, with particular focus on OAuth and SAML-based single sign-on systems.

5. QR code phishing (quishing)

Quishing has evolved from a novelty into a primary attack vector. Attackers embed malicious QR codes in emails, physical documents, parking meters, and even restaurant menus. Scanning the code takes the victim to an AI-generated phishing page optimized for mobile screens.

The attack exploits a fundamental gap: most email security tools do not scan QR code destinations, and mobile browsers offer fewer visual cues about URL legitimacy than desktop browsers. Abnormal Security's 2026 report found that quishing attacks increased by 587% in the first half of 2026 alone.

Mobile extortion apps and crypto scams

Beyond the five primary categories, the advisory also flagged the continued growth of mobile extortion apps (malware disguised as legitimate apps that lock devices and demand payment) and AI-powered crypto scams. The latter use AI-generated deepfake videos of public figures endorsing fraudulent investment platforms. The FTC reported that crypto scam losses reached $5.6 billion in 2025, with AI-generated content playing an increasing role.

Why Traditional Fraud Detection Is Failing

The common thread across all five threat vectors is that AI-generated attacks are designed to look legitimate. They pass the checks that traditional systems rely on:

• Content filters miss AI-written phishing because the text is indistinguishable from genuine communication
• Blocklists miss new infrastructure because attackers rotate domains and IPs continuously
• Identity verification misses synthetic identities because the documents are AI-generated and internally consistent
• Rule-based thresholds miss AI-orchestrated attacks because they are calibrated to stay just below detection limits

Rules-based fraud detection, the kind that checks individual signals against static thresholds, was built for a world where attacks were manual and followed predictable patterns. That world no longer exists.

What Works: Multi-Signal Risk Scoring

The shift required is from evaluating signals in isolation to evaluating them in combination. A VPN IP address is not inherently suspicious. A free email provider is not inherently suspicious. But a signup attempt from a VPN IP, using a disposable email address, with rapid form completion and a device fingerprint seen across multiple accounts? That combination is a strong fraud signal.

This is multi-signal risk scoring, and it is the foundation of effective fraud prevention in 2026.

The signals that matter

No single signal is definitive. The power is in combination.

Implementing Multi-Signal Fraud Detection with Fidro

Here is how to build a layered fraud detection system using Fidro's API. We will check email validity, IP risk, and VPN status, then calculate a composite risk score.

Step 1: Validate the email

curl -X POST https://api.fidro.io/v1/validate/email \
  -H "Authorization: Bearer YOUR_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{"email": "prospect@suspicious-domain.xyz"}'

Response:

{
  "email": "prospect@suspicious-domain.xyz",
  "valid": false,
  "disposable": true,
  "risk_score": 0.91,
  "domain": {
    "name": "suspicious-domain.xyz",
    "mx_valid": true,
    "spf": false,
    "dkim": false,
    "dmarc": false,
    "age_days": 12
  },
  "recommendation": "block"
}

A 12-day-old domain with no SPF, DKIM, or DMARC and flagged as disposable. This alone is a strong signal.

Step 2: Check the IP address

curl -X POST https://api.fidro.io/v1/check/ip \
  -H "Authorization: Bearer YOUR_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{"ip": "185.220.101.42"}'

Response:

{
  "ip": "185.220.101.42",
  "vpn": true,
  "proxy": false,
  "tor": true,
  "datacenter": true,
  "risk_score": 0.97,
  "country": "DE",
  "isp": "Tor Exit Node",
  "recommendation": "block"
}

A Tor exit node. Combined with the disposable email, this signup is almost certainly fraudulent.

Step 3: Composite risk scoring in PHP

<?php

namespace App\Services;

use Illuminate\Support\Facades\Http;
use Illuminate\Support\Facades\Cache;

class FraudDetectionService
{
    private string $apiKey;
    private string $baseUrl = 'https://api.fidro.io/v1';

    public function __construct()
    {
        $this->apiKey = config('services.fidro.key');
    }

    public function assessSignupRisk(string $email, string $ip): array
    {
        // Run email and IP checks in parallel
        $responses = Http::pool(fn ($pool) => [
            $pool->withToken($this->apiKey)
                ->timeout(3)
                ->post("{$this->baseUrl}/validate/email", ['email' => $email]),
            $pool->withToken($this->apiKey)
                ->timeout(3)
                ->post("{$this->baseUrl}/check/ip", ['ip' => $ip]),
        ]);

        $emailData = $responses[0]->successful() ? $responses[0]->json() : null;
        $ipData = $responses[1]->successful() ? $responses[1]->json() : null;

        // Calculate composite score
        $emailRisk = $emailData['risk_score'] ?? 0;
        $ipRisk = $ipData['risk_score'] ?? 0;

        // Weight email signals more heavily (60/40)
        $compositeScore = ($emailRisk * 0.6) + ($ipRisk * 0.4);

        // Hard blocks: disposable email or Tor exit node
        if ($emailData['disposable'] ?? false) {
            return [
                'action' => 'block',
                'reason' => 'disposable_email',
                'score' => $compositeScore,
                'details' => $emailData,
            ];
        }

        if ($ipData['tor'] ?? false) {
            return [
                'action' => 'block',
                'reason' => 'tor_exit_node',
                'score' => $compositeScore,
                'details' => $ipData,
            ];
        }

        // Threshold-based decisions
        if ($compositeScore > 0.8) {
            return ['action' => 'block', 'reason' => 'high_risk', 'score' => $compositeScore];
        }

        if ($compositeScore > 0.5) {
            return ['action' => 'challenge', 'reason' => 'elevated_risk', 'score' => $compositeScore];
        }

        return ['action' => 'allow', 'reason' => 'low_risk', 'score' => $compositeScore];
    }
}

Step 4: Node.js implementation

const axios = require('axios');

class FraudDetection {
  constructor(apiKey) {
    this.apiKey = apiKey;
    this.baseUrl = 'https://api.fidro.io/v1';
    this.headers = {
      Authorization: `Bearer ${apiKey}`,
      'Content-Type': 'application/json'
    };
  }

  async assessSignupRisk(email, ip) {
    // Parallel API calls for speed
    const [emailResult, ipResult] = await Promise.allSettled([
      axios.post(`${this.baseUrl}/validate/email`, { email }, {
        headers: this.headers,
        timeout: 3000
      }),
      axios.post(`${this.baseUrl}/check/ip`, { ip }, {
        headers: this.headers,
        timeout: 3000
      })
    ]);

    const emailData = emailResult.status === 'fulfilled'
      ? emailResult.value.data : null;
    const ipData = ipResult.status === 'fulfilled'
      ? ipResult.value.data : null;

    const emailRisk = emailData?.risk_score || 0;
    const ipRisk = ipData?.risk_score || 0;
    const compositeScore = (emailRisk * 0.6) + (ipRisk * 0.4);

    // Hard block conditions
    if (emailData?.disposable) {
      return { action: 'block', reason: 'disposable_email', score: compositeScore };
    }
    if (ipData?.tor) {
      return { action: 'block', reason: 'tor_exit_node', score: compositeScore };
    }

    // Score-based thresholds
    if (compositeScore > 0.8) return { action: 'block', reason: 'high_risk', score: compositeScore };
    if (compositeScore > 0.5) return { action: 'challenge', reason: 'elevated_risk', score: compositeScore };

    return { action: 'allow', reason: 'low_risk', score: compositeScore };
  }
}

// Usage in Express
const fraud = new FraudDetection(process.env.FIDRO_API_KEY);

app.post('/api/signup', async (req, res, next) => {
  const { email } = req.body;
  const ip = req.headers['x-forwarded-for'] || req.socket.remoteAddress;

  const assessment = await fraud.assessSignupRisk(email, ip);

  if (assessment.action === 'block') {
    return res.status(403).json({
      error: 'Signup blocked due to security policy.',
      code: assessment.reason
    });
  }

  if (assessment.action === 'challenge') {
    // Require additional verification
    req.requiresChallenge = true;
  }

  next();
});

Defending Against Specific AI-Powered Threats

Beyond the general multi-signal approach, here are targeted defences for each threat vector from the advisory.

Against AI-generated phishing

• Verify sender email domains using SPF, DKIM, and DMARC checks before processing inbound emails
• Use Fidro's email validation to verify that reply-to addresses are not disposable or recently created
• Implement link scanning that follows redirects and evaluates destination domains, not just the visible URL

Against synthetic identities

• Check email domain age: legitimate users rarely have email addresses on domains registered within the past 30 days
• Cross-reference the signup IP's geolocation with the claimed address or phone number country code
• Layer device fingerprinting to detect emulators and virtual machines commonly used in synthetic identity creation

Against AiTM attacks

• Implement phishing-resistant authentication (FIDO2/WebAuthn) where possible
• Monitor for session token reuse from different IP addresses or device fingerprints
• Use IP intelligence to detect proxy and relay IPs commonly used in AiTM infrastructure

Against quishing

• Scan QR code destinations in inbound emails and documents before rendering them
• Educate users to verify URLs after scanning QR codes, especially on mobile
• Implement mobile-specific fraud detection that accounts for the reduced security context of mobile browsers

What This Means for Your Roadmap

If you are building or maintaining a fraud prevention system, the Google advisory demands a re-evaluation of your stack. Here is a practical assessment framework.

Audit your current defences. Map each of the five threat vectors against your existing controls. Where are you relying solely on rules-based detection? Those are your gaps.

Prioritize multi-signal integration. If you are not already combining email, IP, and behavioral signals into a composite risk score, that is your highest-impact next step. Fidro's API gives you email and IP signals in a single integration. Check the API docs and pricing to scope the effort.

Test against AI-generated attacks. Generate test phishing emails using an LLM, create test accounts with disposable emails through VPNs, and verify that your system catches them. If it does not, you know what needs to change.

Monitor the threat landscape. Google's advisory will not be the last. Subscribe to threat intelligence feeds, follow CISA advisories, and track the fraud prevention research coming out of Fidro's blog and the broader security community.

Comparing Fraud Prevention Approaches for 2026

For a detailed comparison of fraud prevention tools and how they stack up against these threats, see our fraud prevention comparison guide.

Key Takeaways

Google's June 2026 fraud advisory makes one thing clear: AI has permanently changed the fraud landscape. The 1,210% increase in AI-powered scams is not a spike that will correct itself. It is the new baseline.

For developers and engineering teams, the response is not panic but pragmatism. Multi-signal risk scoring, layered API-based detection, and continuous monitoring are the foundations. The tools exist. The data is clear. The only question is how quickly you implement them.

Start here:

• Email checker tool for real-time email validation
• IP checker tool for IP risk assessment
• VPN detector for proxy and VPN detection
• API documentation for integration details
• Pricing for volume plans
• Fraud prevention use cases for industry-specific patterns