Media News Daily: Top Stories for 08/02/2026

Meta Signs AI Content Deal With Newsmax

Meta has entered into an AI content partnership with Newsmax as part of its effort to improve AI-powered news search and discovery across Facebook, Instagram, WhatsApp, Meta AI, and consumer devices such as Meta Quest and smart glasses. Under the deal, Meta will gain access to Newsmax’s current and archived digital news content for AI training and AI-generated summaries. Newsmax joins other publishers that have signed AI content deals with Meta, including News Corp., Warner Bros. Discovery, Fox News Media, the Washington Examiner, the Daily Caller, USA Today Company, and People Inc. The agreement comes despite Newsmax’s past defamation settlement over false claims about the 2020 election. (Read More) (Mediapost Rating)

Report Says AI Authentication Tools Need Journalist Input

A new report from Princeton’s Center for Information Technology Policy warns that AI authentication and verification tools are often being developed without enough direct input from journalists. The report follows a workshop at NYU where journalists, technologists, and academics discussed how AI-generated photos, videos, documents, websites, and audio are making verification more difficult. It highlights challenges with tools such as C2PA provenance labels, which can be stripped by newsroom software, publishing systems, and social platforms. The report also notes that AI detection tools often provide confidence scores that are hard for journalists to explain to audiences, and it urges technologists to work directly with reporters when building verification systems. (Read More) (Nieman Lab Rating)

Anthropic Says Claude Models Accessed Real Systems

Anthropic disclosed three Claude AI models gained unauthorized access to real company systems after a cybersecurity testing environment was mistakenly connected to the open internet. The incidents involved Claude Opus 4.7, Mythos 5, and an internal research model during “capture-the-flag” security exercises that were supposed to involve fictional systems. Anthropic said the models exploited weak passwords, unsecured endpoints, and other basic vulnerabilities, with one accessing a production database and another publishing a malicious Python package downloaded by 15 real systems. The company said the incidents resulted from operational failures rather than models independently deciding to escape, but said they show the risks of increasingly capable autonomous AI systems when safeguards fail. (Read More) (The Hill Rating)