The beautiful thing about AI and robotics is that you're never done."
— Manuela Veloso
Supriya Singh is a Reporter at AI FrontPage covering the AI & Education and AI & Jobs beats. She brings six years of print and digital experience, including three years at The Asian Age, where she reported on higher education, Delhi government, and crime. She is based in Delhi-NCR.
AI models don’t just invent fictional names at random, they keep reaching for the same ones. A new preprint finds that “ghost” identities like Elena Vasquez and Marcus Chen recur across hundreds of AI-generated documents, fake therapy clinics and even 1,655 backdated records on a CERN-run research repository.
While watermarking is viewed as a safeguard for tracing AI-generated content, research presented at ICML 2026 shows that building watermarks that are secure, robust and stealthy at the same time remains a fundamental challenge.
The International Conference on Machine Learning (ICML) 2026 has announced its awards, recognising two outstanding papers, five honorable mentions, two position papers and a Test of Time Award. Diffusion models dominated the top honours, while a decade-old reinforcement learning paper — one that quietly shaped how today’s large language models are trained ,was named the year’s most enduring contribution.
Beyond the corporate scorecard, the numbers point to a quieter shift: open model families are becoming the default foundation on which academic AI research is now built, with one chip-maker sitting squarely at the centre of it.
Large language models often change the direction of social media posts on controversial topics even when explicitly instructed to preserve the original meaning, an Oxford-led study finds.
The world’s premier machine learning conference opens in Seoul with record-breaking numbers — 23,918 paper submissions, more than double last year — and an unprecedented experiment: formal rules on whether AI can help review AI research, enforced by watermarks hidden inside the papers themselves.
According to the survey by UNICEF, which is based on data from 10 countries, at least 20 million children have used AI and many are outpacing adults by adopting it at rates more than three times faster.
Backed by an investment of $2.5 billion the company will provide a unique combination of skills inclusive of deep industry knowledge, change management and continuous improvement experience, and enterprise-grade AI engineering expertise.
“A country is not sovereign because it owns every technology it uses. It is sovereign when it is not dependent on a single provider, a single geography or a single political decision”, says Amitabh Kant in an exclusive interview with AI FrontPage.
The 40-member Independent International Scientific Panel on AI — co-chaired by Yoshua Bengio and Maria Ressa — released its first report on July 1, warning that AI safeguards are failing to keep pace with capabilities and that the Global South remains locked out of both AI development and governance. Guterres to governments: “Do not wait.”
