OpenAI Embeds Persistent Watermarks in GPT‑5 Outputs to Thwart Deepfakes

OpenAI announced that its forthcoming GPT‑5 model will automatically embed a covert watermark into any text, image, or audio it generates. The watermark is engineered to survive common post‑processing steps such as compression, resizing, or format conversion, and can be detected using a free, open‑source toolkit that verifies the hidden signal.

The built‑in watermark gives content platforms a reliable way to flag AI‑generated media, making it harder for malicious actors to circulate convincing deepfakes at scale. For defenders, this provides an actionable indicator that can be integrated into detection pipelines, threat‑intel workflows, and incident‑response playbooks, helping to attribute malicious content and reduce the time spent on manual analysis. However, adversaries may attempt to strip or obscure the watermark, so monitoring for both the presence and potential tampering of these signals will be essential.

Categories: AI Security & Threats, #AI Security & Threats

Source: Read original article