All articles tagged with #ai training
Anthropic has introduced a new optional setting for Claude users to share voice recordings for AI model training. The feature, which appears as a prompt during voice interactions, is disabled by default and separate from existing text-based training permissions. Users can enable or disable the option in privacy settings and delete data at any time.
A 404 Media investigation reports that OpenAI’s contractor-based process for reviewing and training ChatGPT data has led to several contractors being fired for using AI themselves. Internal guidelines reportedly prohibit workers from using AI tools during evaluation, yet some sources say people were terminated for AI use, highlighting the irony of AI being used to train the AI while its human evaluators face penalties. OpenAI did not comment on the firings.
A WIRED review of Meta's Muse finds the personal AI assistant offers real convenience—web browsing, shopping, and task automation—but nudges users to share extensive data (emails, bank details, documents) and stores interactions in a 'Memory' that can't currently be fully disabled. Data can be used for AI training, even if sanitized, raising privacy concerns and potential influence on ads. While Meta defends the defaults as improving the model, critics warn of data exploitation, and the author ultimately deleted Muse after testing.
The Trump Administration filed a DOJ brief backing OpenAI in The New York Times’ copyright lawsuit, arguing that training AI on copyrighted articles is extraordinarily transformative and likely fair use, a stance the government says could safeguard AI leadership and innovation in the U.S. while not binding on the judge; the filing comes amid broader AI copyright battles and drew mixed reactions from The Times, the Authors Guild, and others.
Sony Music Publishing and Warner Chappell filed a California lawsuit accusing Anthropic and its executives of scraping, torrenting and downloading tens of thousands of copyrighted songs and lyrics to train Claude AI, seeking billions in damages and alleging data stripping to deny attribution; the case cites hits like All I Want for Christmas Is You and Eye of the Tiger and follows a related $1.5B settlement in a separate copyright dispute against Anthropic.
Mo Zohourian moved from Iran to Canada and, after months of unsuccessful job hunting, started AI training work at about $15/hour. Through passing assessments (English literature, reasoning, mathematics) and leveraging his sales background, he climbed to $30–$70/hour on various projects. When expert, sales-focused opportunities emerged, he secured higher-paying, consistent work and has earned around $100/hour as an AI evaluation and sales-management specialist for the past 10 months. He also founded Annotation Academy and an AI consulting firm for small businesses, works 20–40 hours weekly with flexible hours, reinvests earnings into his ventures, and enjoys time with his son—the result of a booming demand for domain expertise in AI.
Sony Music and Warner Music have filed a federal lawsuit in Northern California accusing Anthropic of illegally scraping tens of thousands of copyrighted songs to train its Claude AI, seeking damages and signaling a broad escalation in AI copyright battles.
A Twitch streamer has filed a proposed class action in the Northern District of California accusing Twitch and Amazon of copying vast quantities of streamer content—including broadcasts, clips, chat logs and photos—to train Amazon's generative AI without licenses or consent. The suit cites breach of implied contracts and unfair competition. Twitch had announced plans to train AI on platform content with an opt-out that is not retroactive and is tied to individual channels. Amazon acquired Twitch in 2014; the case underscores growing legal scrutiny of AI training on creator content.
Connecticut Twitch streamer Warren Pandiscia leads a proposed class action accusing Twitch and its parent Amazon of scraping millions of streams to train Amazon’s generative AI without permission, despite Twitch’s opt-out option. Plaintiffs allege breach of implied/express contracts and unfair competition and seek injunctive relief, damages and disgorgement. The suit, filed in the Northern District of California, follows Twitch’s update of its terms to allow AI training with an opt-out by channel, noting that content from opted-out users can still be used if they engage on opted-in channels. No immediate comment from the parties.
A hidden AirTag found in a shipment of rare books is said to have led investigators to an Amazon facility where books are sliced from their bindings and scanned for use in AI training, supporting long-standing suspicions that some bulk used-book orders are intended to feed AI systems without author consent or payment; the 404 Media investigation—which is paywalled—alleges this practice is underway and highlights legal and IP gaps as laws struggle to keep pace with AI-enabled text collection.
An investigation traces a rare book to Amazon’s Las Vegas AI training facility where staff reportedly dismantle and scan volumes to train AI models, prompting concerns that irreplaceable copies could be destroyed and copyright issues may arise, even as Amazon says purchases are used to improve products.
A 404 Media investigation tracked a shipment of a scarce, pre-ISBN book to Amazon’s LAS8 facility in Las Vegas and uncovered a separate operation, VGT3, where workers cut spines and scan pages—likely to train AI models—tracked by an AirTag inserted into the book. Amazon says it buys books to develop its products, but hasn’t confirmed the use, and the piece places the finding in a broader history of investigative tracking and a 2025 fair-use ruling on digitizing bound volumes.
A 404 Media investigation, aided by an Airtag placed in a rare book, traces bulk orders of rare volumes to an Amazon AI training facility in Las Vegas where staff reportedly tear books from their spines to scan pages for frontier AI models; Amazon says it buys books to develop products, but critics warn that destroying rare texts raises ethical concerns and could erode access to valuable literature.
Twitch has added an option for users to opt out of Amazon using their streams, VODs, clips, and chat on Twitch to train Amazon’s generative AI models. Previously, users appeared to be auto-enrolled in AI training. Opt-out requires adjusting settings at twitch.tv/settings/security, and opting in can contribute to improvements in AI such as speech-to-text and captions. The move increases user control and transparency, though some users remain skeptical about Amazon leveraging their content for AI development.
Germany’s Munich Regional Court has ruled that Suno infringed GEMA’s copyrights by using and training its AI on protected songs without licenses, ordering Suno to disclose revenue and pay damages (amount TBD); the decision is a first-instance ruling and can be appealed. GEMA, which represents about 95,000 members, argued the AI-generated content matched original melodies, while Suno says its system creates new songs and plans to appeal.