This is AI Frontiers. Today, why real people make fake stories convincing? A genuine signal proves far less than it seems. The mistake is granting one true piece of evidence authority over a story it cannot prove. Anthropics' September 2026 Threat Report describes a deceptive dating app network that combined artificial intelligence personas with paid human workers. Over two weeks in April, it observed more than 4,700 AI personas interacting with at least 25,000 people. Real workers supplied live video calls and social media followbacks that helped reassure users. Even those workers were AI assisted. A separate model proposed their replies and they tapped one. These are one provider's findings from one operation, not market-wide estimates. The work can be real too. Microsoft's March 2026 research describes North Korean information technology worker operations using AI to sustain everyday correspondence, technical responses, and coding work under fraudulent identities. Microsoft says AI helps actors, quote, get hired, stay hired, and misuse access at scale. In the Christina Chapman case, the United States Department of Justice reported 309 U.S. businesses defrauded and more than $17 million generated by a remote worker scheme supported by stolen identities and a domestic laptop farm. It ran from 2020 to 2023. Those figures are not a measure of losses caused by generative AI. Agreement between sources is only worth as much as their independence. Four documents supplied by the same party are not four confirmations. A reference reached through contact details obtained from the former employer tests a different relationship from one reached through the applicant's preferred address. Detection still matters, but its result has a boundary. A September 2026 preprint introducing DF26 found participants correctly identified 52.6% of its fake clips, compared with 74.5% on CLib DF and 69.8% on DeepSpeak V2. Five-second clips, no audio, mostly technical university volunteers. A separate March 2026 preprint found mean human accuracy of 78.4% versus 53.7% for the tested detector variants. Different tasks, different samples. There is no useful universal score. Cloudflare's 2026 announcement with name tag adds identity verification to workplace access. A selfie and government-issued identity document produce a token used in the access decision. That is a workflow, not proof of honest purpose. So ask one question in every hiring review or access decision. What would still be unproven if every item in front of us were genuine? Then narrow the access or contact the organization yourself. That's AI Frontiers.