AI This Week

A new report highlights how Google’s AI Overviews feature is driving a sharp increase in “zero clicks” for news-related searches, leaving publishers struggling. Since its launch in May 2024, the tool has contributed to a drop in traffic to news sites — a decrease from 2.3 billion visits in mid-2024 to 1.7 billion in May 2025, as per SimilarWeb data. AI summaries allow users to consume information directly, bypassing traditional sources. Google has faced criticism but argues its AI tools create new opportunities for content discovery. Meanwhile, referral traffic from ChatGPT has surged, offering a potential lifeline for some outlets. As the media industry faces layoffs and shifting referral patterns, questions remain about how publishers can adapt.

A new initiative is bringing artificial intelligence into classrooms across the United States. Anthropic, Microsoft, and OpenAI are funding the National Academy for AI Instruction, a $23 million project aimed at training educators on using AI technologies. The American Federation of Teachers is spearheading this effort, which will offer free virtual training to its 1.8 million members. The academy, set to launch this fall in Manhattan, plans to train 400,000 educators over five years.

Wimbledon faced major embarrassment during Sonay Kartal’s match against Anastasia Pavlyuchenkova when an official accidentally deactivated the AI line-judging system mid-game. The incident left critical calls unaddressed, forcing chair umpire Nico Helwerth to halt play. With the score tied and tensions high, players were left uncertain in what proved to be an embarrassing moment for tournament organizers. Former champion Pat Cash criticized the umpire’s inability to act, calling the situation “ridiculous.”

The All England Club admitted the mishap stemmed from human error and reviewed its processes afterward. This AI glitch comes amid rising player complaints about inaccuracies, raising concerns over its efficacy, especially on grass courts. Questions persist about Wimbledon’s reliance on technology without backup video review systems.

Elon Musk’s artificial intelligence startup, xAI, has raised a staggering $10 billion in combined debt and equity. Morgan Stanley confirmed that $5 billion came from secured notes and loans, while the other $5 billion resulted from strategic equity investments. These funds will drive the development of xAI's flagship Grok AI model and bolster infrastructure, including its Colossus supercomputer in Memphis—a facility with plans to amass 1 million GPUs. Positioned as a rival to OpenAI and Anthropic, xAI aims to disrupt the competitive AI landscape with its distinct "truth-seeking," controversial approach. Musk has integrated Grok with the X social media platform and continues scaling operations. The firm’s valuation, once at $80 billion, may evolve with this new funding.

A coalition of top cybersecurity agencies, including NSA, FBI, and CISA, has unveiled new guidance to combat emerging threats to AI systems. The guidance warns about risks like data poisoning, supply chain flaws, and data drift that could compromise AI systems. It urges organizations to implement security measures across all phases of the AI lifecycle—planning, data collection, model building, and operational monitoring. Key recommendations include verifying third-party datasets, using secure ingestion protocols, and auditing system behavior regularly. The guidance stresses preventing model poisoning and tracking data lineage for integrity. Organizations are encouraged to revise their incident response plans, audit ongoing projects, and build cross-functional teams. As AI models become integral to critical infrastructure, robust data security is essential.

Amazon recently unveiled two major advancements in its automation strategy. The retail giant has deployed its one millionth robot in a fulfillment center in Japan, marking a notable leap in robotic infrastructure. Alongside this achievement, the company introduced "DeepFleet," a generative AI model designed to enhance the efficiency of its robotic fleet by 10%. DeepFleet functions as an advanced traffic management system, optimizing robotic movements across fulfillment centers while leveraging AWS tools like Amazon SageMaker. The system continuously improves through AI-driven learning, with promises of greater logistical efficiency. Notably, this development aligns with Amazon’s broader push for generative AI, despite concerns over potential job impacts.

Microsoft AI has unveiled groundbreaking research that could reshape medical diagnostics. Through the Sequential Diagnosis Benchmark (SDBench), the company tested AI against 304 notoriously complex medical cases from the New England Journal of Medicine. The model-agnostic diagnostic tool, MAI Diagnostic Orchestrator (MAI-DxO), demonstrated 85.5% accuracy, surpassing the average performance of human physicians. By expediting decision-making and reducing costly tests, this innovation offers both precision and cost-effectiveness. Microsoft believes this may be a step toward "medical superintelligence," described as a system capable of outperforming the collective expertise of global clinicians. While limitations remain, the research signals major implications for health care's future.

Baidu is set to release its Ernie generative AI large language model to the open-source market, marking China's most significant AI development since DeepSeek. The rollout signals a shift from Baidu’s traditional proprietary strategy and intensifies competition among leading global tech firms. Experts suggest the move could disrupt pricing dynamics, pushing competitors like OpenAI and Anthropic to reassess their gated premium models. Industry leaders are divided on whether Ernie will rival DeepSeek’s impact, but many agree its open-source nature raises the stakes for innovation and accessibility. Questions surrounding security, transparency, and geopolitical implications also loom large as Baidu positions its AI tools for widespread adoption.

A recent study from the Max Planck Institute reveals that people, especially academics and professionals, have begun adopting AI-like language patterns in their daily speech. By analyzing 280,000 academic YouTube videos, researchers identified a significant increase in the use of terms commonly generated by AI, such as “meticulous” and “adept.” This linguistic shift might erode emotional nuance and individuality, reducing speech to monotony. The study warns that overusing AI-influenced language could flatten cultural diversity and subtly alter social behaviors, like politeness in conversations. The irony? Humans created AI to sound human, but now people are mirroring AI.

Washington University School of Medicine in St. Louis has developed an AI-based brain mapping technology that has received FDA authorization for use in neurosurgery. The Cirrus Resting State fMRI Software rapidly identifies critical brain areas controlling functions like speech and movement, enhancing precision in tumor removal and epilepsy treatment. This technology, licensed to startup Sora Neuroscience, analyzes brain activity patterns in just 12 minutes of fMRI scanning. It offers broader accessibility for brain mapping, including for patients who struggle with traditional task-based fMRI