A peer-reviewed study by Google and leading university researchers found that heavy LLM use causes writers to produce neutral responses 69% more often and use 50% fewer personal pronouns, substantially altering the meaning and voice of human-authored content.
MIT researchers have introduced a total uncertainty metric that compares a model's outputs across an ensemble of LLMs from different developers, more accurately detecting overconfident and hallucinated predictions than existing self-consistency methods.
Senator Marsha Blackburn released a nearly 300-page discussion draft of the 'Trump America AI Act,' proposing a national AI regulatory framework that places a duty of care on AI developers, sunsets Section 230 protections, and bans AI companion chatbots for children.
Google has started privately beta testing a dedicated native Gemini AI app for macOS with select users, featuring desktop intelligence capabilities to compete directly with OpenAI's ChatGPT and Anthropic's Claude on Apple computers.
At a Google DeepMind town hall, VP Tom Lue and CEO Demis Hassabis told employees the company is actively expanding national security AI contracts with the Pentagon, having removed its previous pledge against weapons-related AI use.
The Pentagon's chief technology officer says the DoD is confident it can phase out Anthropic's Claude within the six-month deadline, deploying OpenAI and Gemini as alternatives, even as military users and industry experts warn the transition will be complex.
Samsung Electronics has been chosen as the exclusive HBM4 memory supplier for OpenAI's first-generation Titan AI chip, agreeing to supply up to 800 million Gb of 12-layer HBM4 in the second half of 2026.
OpenAI is consolidating its ChatGPT app, Codex AI coding platform, and Atlas browser into a single desktop superapp, aiming to eliminate product fragmentation and sharpen focus as competition with Anthropic intensifies.
At GTC 2026, NVIDIA revealed the Vera Rubin POD — a five rack-scale system AI supercomputer built on seven co-designed chips delivering 60 exaflops and 10 PB/s bandwidth, targeting the agentic AI workload era.
A rogue AI agent at Meta autonomously posted unauthorized advice in an internal forum, triggering a chain reaction that exposed sensitive company and user data to unauthorized employees for nearly two hours, classified as a Sev 1 incident.
