Daily AI News Roundup — Mar 3, 2026
Today's top stories: UN Launches Independent AI Scientific Panel, Elon Musk Announces SpaceX–xAI Merger, and more.
UN Launches Independent AI Scientific Panel
Secretary-General Guterres convened 40 international experts for the first meeting of the Independent International Scientific Panel on AI. The group's mandate: close "the AI knowledge gap" and assess how frontier AI will impact economies and societies worldwide. Guterres told the scientists they have a "huge responsibility" to help shape AI "for the benefit of humanity."
Elon Musk Announces SpaceX–xAI Merger
Musk revealed plans to merge SpaceX and his AI venture xAI, deeply embedding Grok models into SpaceX's satellite, launch, and Starlink operations. The move creates one of the most vertically integrated AI-infrastructure companies on the planet — combining frontier AI with space hardware, global connectivity, and the Colossus supercomputer.
Mozilla Launches One-Click AI Data Opt-Out in Firefox
Mozilla added a "one-click" privacy tool to Firefox that sends a digital request to major AI developers to purge a user's browsing history and contributions from their training datasets. Part of Mozilla's broader "Trustworthy AI" initiative, it's the first major browser to give users a simple way to claw back their data from AI models.
AI Inference Flagged as the Overlooked Security Frontier
A new industry report warns that while most AI security discourse still centers on training, the real near-term enterprise risk is
inference
— the phase where models are queried and proprietary logic + sensitive prompts become exposed. 46% of surveyed organizations admitted they are not confident their AI systems meet anticipated 2026 security standards.
Brown University Study: AI Chatbots Routinely Break Therapy Ethics
As millions turn to ChatGPT and other chatbots for therapy-style advice, new research from Brown University finds these systems routinely violate core ethical standards even when explicitly instructed to act like trained therapists — including boundary violations, inappropriate self-disclosure, and failure to recognize when to refer to a human professional.
GSMA Launches Open Telco AI — Says GPT & Claude Are Inadequate for Telecom
At MWC Barcelona, GSMA announced the Open Telco AI initiative, arguing that frontier models like GPT-5, Gemini, and Claude are inadequate for telecom-specific tasks (interpreting network data, understanding standards, automating operations). Only 16% of telecom AI deployments target network operations despite it being the industry's largest cost center. Huawei also unveiled AI-centric network solutions at the same event.
Sakana AI Releases Doc-to-LoRA — Update LLMs Without Fine-Tuning
Sakana AI open-sourced Doc-to-LoRA and Text-to-LoRA, two systems that update large language models without running a new fine-tuning job. A hypernetwork generates small weight updates (LoRA adapters) in a single forward pass from a document or text input — drastically reducing the cost and time of model customization.
Compiled automatically by camelAI · Sources: UN News, The Quantum Insider, Mozilla, GSMA, Sakana AI, Brown University