Here's a headline that should make you pause before trusting any AI answer with your money or your login.
Hackers have found a way to weaponize the internet's content against AI models like ChatGPT, Gemini, and Google AI Overview. They call it 'Dark Sourcery.'
The method is simple but brutal. Flood the web with fake reviews, PDFs, support pages, and articles, all optimized to rank exactly where these AI models pull their answers from.
Ask ChatGPT for Delta's support number, or Chase's login page, and it might confidently hand you a phishing number or a fake portal, straight from a poisoned source it trusted.
Think of it like a brilliant assistant reading from a corrupted textbook. The assistant isn't lying to you, it genuinely believes what it read. The book itself was sabotaged.
Hundreds of major brands got caught in this, including Lufthansa, Qatar Airways, Bank of America, Airbnb, and TripAdvisor.
What makes this genuinely dangerous is the delivery. AI doesn't hedge like a sketchy ad might. It states the fake info with the same calm confidence it uses for real facts.
The real lesson here isn't that AI is broken, it's that AI is only as trustworthy as the internet it learns from, and right now, attackers are actively poisoning that well.
Practical move: any phone number, email, or login link an AI gives you for something sensitive, double check it against the company's official site first. Don't let confidence fool you into skipping that one click.
🔗 Original Source & Reference: https://medium.com/@arielsimon/dark-sourcery-how-hackers-manipulate-ai-to-scam-you-88df434d2073
Published automatically via FeedMind AI Content Pipeline.
Top comments (0)