Fake AI Ad Portals Stealing Credentials: What LLM App Builders Need to Know

The Threat: Fake AI Ad Portals Are Harvesting Credentials at Scale

Cybersecurity researchers have uncovered a sophisticated phishing operation targeting users of major AI platforms. The attackers created convincing fake advertising portals for ChatGPT, Google Gemini, Anthropic Claude, Perplexity, Meta Muse, and Manus—claiming to offer legitimate services like campaign optimization, spend audits, and business-account management.

The real objective? Harvesting user credentials and multi-factor authentication (MFA) codes. According to The Hacker News, this "human-operated phishing platform" demonstrates a level of sophistication that should concern both individual users and organizations building on LLM infrastructure.

Why This Matters for LLM App Builders

This attack vector reveals critical vulnerabilities in how AI platform users interact with authentication systems. For builders and companies developing applications powered by ChatGPT, Claude, Gemini, and similar models, the implications are serious:

• API Key Exposure: Compromised credentials could grant attackers access to your API keys, allowing them to make unauthorized requests and incur unexpected costs.
• Data Breach Risk: If users reuse passwords, attackers gain access to multiple systems and sensitive data associated with your LLM application.
• Trust Erosion: Your users may lose confidence in your platform if their AI service credentials are compromised through phishing.
• Regulatory Liability: Depending on your jurisdiction and data handling practices, credential breaches can trigger compliance violations.

The Guardrail Problem

Traditional security guardrails—like warning users about suspicious domains or implementing certificate pinning—are essential but insufficient against this threat. The phishing portals were designed to look nearly identical to legitimate ad dashboards, making visual inspection ineffective.

This highlights a critical gap: most LLM platforms lack built-in credential verification warnings at the application layer. Users are often responsible for identifying phishing attempts, which is unreliable at scale. Builders relying on ChatGPT, Claude, or Gemini APIs have limited control over how their users authenticate with these platforms.

What Builders Should Do Now

1. Educate Your Users

Create clear documentation about secure authentication practices. Warn users against entering API keys or credentials into unfamiliar portals, and advise them to access official dashboards only through bookmarks or official links.

2. Implement Application-Level Protections

• Use OAuth 2.0 or similar delegation protocols when possible, reducing the need to handle raw credentials.
• Implement API rate limiting and monitoring to detect unusual activity that might indicate compromised keys.
• Add logging and alerting for authentication failures and suspicious access patterns.

3. Strengthen Your Own Authentication

If your LLM app requires users to authenticate, ensure you're not asking for API keys directly. Instead, use official SDK authentication flows provided by OpenAI, Anthropic, and Google.

4. Monitor for Threats

Subscribe to security advisories from your LLM providers and implement intrusion detection systems that flag unusual API usage patterns.

5. Enforce MFA in Your Own Systems

Since attackers specifically target MFA codes, require MFA for all accounts accessing your LLM application, and educate users about MFA security best practices.

The Bottom Line

This phishing campaign exposes a critical vulnerability in the AI application supply chain: users are the weakest link in credential security. As a builder, you cannot rely on end-users to identify phishing attacks. Instead, architect your application to minimize credential exposure, use delegation protocols wherever possible, and implement robust monitoring systems.

The threat is real, and it's active now. The time to strengthen your security posture isn't after an incident—it's today.