AI vs. the Fakes: Inside the 2025 Race to Spot ChatGPT, Deepfakes and More
OpenAI discontinued its AI text detector in 2023 due to low accuracy, while Turnitin's tool misidentified over half of essays in a Washington Post test. The EU's AI Act will require disclosure and watermarking of AI-generated content by 2026. Researchers in 2024 removed or faked watermarks with 85% success. Voice phishing attacks surged 442% in late 2024 as AI-generated voices and videos spread.