Press Gazette launches Made by Humans to screen newsroom content for AI fakes

Press Gazette has launched Made by Humans, a pre-publication verification tool intended to help newsrooms identify AI-generated or manipulated material before it is published. The service is available to publishers and individual journalists, and assesses text, images, video and audio. Press Gazette devised the product with Dublin-based Flip-Pay, which built it from a concept developed by Press Gazette editor-in-chief Dominic Ponsford. It is designed to address the editorial, legal and reputational risks of publishing misleading AI-produced content, while supporting rather than replacing editorial judgment. The platform combines Pangram's text analysis with Identifai's media checks in a single dashboard. It can be integrated into WordPress and other content-management systems, where it warns editors before publication if material shows signs of AI manipulation. It also keeps a record of reviews and interactions for compliance and audit purposes. Editors can assign material one of three classifications: “AI Generated,” “AI Assisted” or “Human Made.” The tool can check whether material is original, plagiarised or has appeared elsewhere, and supports English, French, Spanish and German at article and sentence level. Publishers can also create reports and seek a trust badge intended to signal responsible AI use to readers and advertisers. Press Gazette first trialled the service in its own content-management system in September. Flip-Pay said it was available on a pay-as-you-use basis through madebyhumans.press. The launch follows Press Gazette's Reality Wars investigations, which identified more than 1,000 examples of mainstream-media articles citing fake or AI-generated experts. The investigations also exposed Margaux Blanchard, a fake freelance journalist who used AI to pitch invented news features that were published. In another case, newspapers internationally were deceived by an AI-manipulated image posted by a Thai police force, which falsely claimed detectives had dressed as showgirls for an undercover drugs bust. A Press Gazette survey of more than 200 publishers, editors and journalists, conducted in August, found that 95% believed AI-generated content should be flagged to readers. About 64% said their publication had no reliable way to detect unauthorised AI use, while 87% thought a human-made guarantee would improve reader trust. The component suppliers make their own performance claims. Identifai says its lead image model detects AI-generated images with 99.8% accuracy in its benchmarking, and that its cloned- and synthetic-voice model reaches 91.8%. Pangram says its fourth-generation model produces one false positive for every 10,000 detections, and can identify text generated by OpenAI, Anthropic, Google, Meta and DeepSeek. Identifai logged about 10,000 deepfake incidents reported in English-language media or on social platforms since January 2020. Its tally rose from four incidents in January 2020 to more than 3,000 in March 2026. The broader verification problem is recognised in joint guidance by cybersecurity agencies in the United States, Australia, Canada and Britain. The guidance says AI-powered manipulation has made media verification more difficult and can enable impersonation, misinformation and other malicious activity that erodes trust in online media. It distinguishes such detection from content provenance: cryptographically signed Content Credentials can document a media item's source and editing history, but do not themselves establish that the content is true.