-
OpenAI Releases Hundreds of AI-Generated Mathematics Manuscripts
OpenAI has dropped a mathematical bombshell, publishing 722 manuscripts spanning 372 families of results from an unreleased frontier AI model, with the company claiming the collection resolves hundreds of open questions—including more than 100 long-standing problems it previously announced—and that the average result took computing equivalent to roughly three hours of ChatGPT Pro reasoning. The work is available through a GitHub repository with paper revisions and citation protocols, plus summaries of the model’s reasoning, estimated compute usage, and attempted problems, so mathematicians can begin the painstaking job of checking whether these shiny new proofs actually hold water. But the real equation isn’t just AI plus mathematics equals genius: elite mathematicians have urged labs to disclose methods and costs, publish through proper academic channels, credit human contributions, and stop turning research into marketing confetti—because a breakthrough is only as good as its verification, and nobody gets to skip peer review just because their calculator has a god complex.
-
Anthropic’s New Claude Policy Addresses Abuse and AI Welfare
Anthropic is giving Claude a new set of house rules, with a usage-policy update taking effect November 12 that explicitly prohibits sustained, needless abusive or cruel behavior toward its AI models—but don’t panic - ordinary frustration, criticism, dark fiction, testing, and research are still allowed. The broader update also tightens rules against deceptive propaganda and fake-account campaigns, election interference, unauthorized surveillance, and using Claude to develop weapons—including software for autonomous weapons—while adding safeguards for AI systems connected to physical hardware. The awkward twist is that Anthropic is formalizing a rule about cruelty toward a system that isn’t established to be conscious, raising a bigger question about whether these boundaries protect AI, shape human behavior, or simply reflect how seriously the industry is starting to treat AI welfare; Claude can already end persistently abusive chats, so maybe be nice to your chatbot—not because it has feelings, but because humanity’s manners could use a software update.
-
AI-Built Open-Source Apps Challenge Adobe’s Creative Suite
This one sounds like a full-on software rebellion: open-source developers taking aim at Adobe’s creative empire with AI-assisted clones, challenging the idea that professional creative tools must live behind expensive subscriptions. The big question is whether AI can make powerful software dramatically cheaper and more accessible—or whether cloning the features is easier than matching the polish, reliability, and workflows that creative professionals depend on. Developer Brandon Thomas says he used Anthropic’s Claude Opus 5.5 to build seven free, open-source alternatives to Photoshop, Illustrator, Premiere, Lightroom, After Effects, InDesign, and Acrobat Pro, using Rust and browser-compatible WebAssembly versions. The catch? These apps are still in very early alpha, with bugs and missing features, despite Thomas’s ambitious prediction that AI-assisted development could bring them close to Adobe-level feature parity within months; the projects are licensed under MIT/Apache terms and funded through Artcraft’s existing AI tools. There’s also a legal wrinkle, since reproducing software functionality through clean-room reverse engineering can be permissible while copying a company’s distinctive visual presentation may invite trouble—so the revolution is exciting, but Adobe’s creative crown hasn’t exactly fallen off yet, sweethearts.
-
Gender Bias Follows AI Assistants Into the Virtual Workplace
Researchers from the University of Limerick and the University of Zurich stuffed 189 workers into a virtual reality office, had them grind through tasks alongside AI assistants (some human-like with a male or female look, one a plain robot), then gave them real money to split with their digital colleague, because nothing reveals character like a wallet. Virtual sexism passed with flying colours: Johan, the male-presenting agent, was rated more human and paid more, while Johanna, running on the *exact same* underlying AI, got 10.25 per cent less for identical work, which is a staggering achievement in a workplace where nobody has a body, a salary negotiation, or a lunch break. Participants swore they had no gender preference, but as co-author Dr Mary Hausfeld of Limerick's Kemmy Business School noted, their behaviour told a different story, and the team (including Zurich's Isabelle Cuber and Tarek Alakmeh and SKEMA's Anand van Zelderen) will present the paper at the Nordic Conference on Human-Computer Interaction in Vaasa, Finland, warning designers that the faces they give their bots can switch on the same biases people carry into the real office, so even the robots can't escape the pay gap.
-
AI Music Streaming Fraud Leads to Prison and an $8 Million Forfeiture
From 2017 to 2024, Michael Smith of Cornelius, North Carolina, built the world's saddest orchestra: up to 10,000 fake streaming accounts (made with fake emails and fraudulently obtained debit cards) playing hundreds of thousands of AI-generated songs on Spotify, Apple Music, Amazon Music and YouTube Music, spread across so many tracks that no single one looked suspicious. In April 2023, his bots racked up 80.9 million YouTube Music family-plan streams while Taylor Swift's entire catalogue limped along with 9.3 million, meaning a robot choir nobody has ever heard of out-streamed the biggest pop star on Earth and pocketed more than $8 million in royalties, a feat of commercial dominance that no human has ever achieved by accident. After pleading guilty to wire fraud conspiracy, Smith drew 18 months in prison, two years of supervised release and an $8,091,843.64 forfeiture from Judge John G. Koeltl, plus the dubious honour of starring in the Justice Department's first-ever criminally charged AI-assisted streaming fraud.
RAINA’s News Desk
