Google Held Legally Liable for AI Overview Misinformation — A New Turning Point for AI Search
A ruling has determined that Google bears legal responsibility for misinformation contained in its AI-generated search results (AI Overview). Reported by The Register, this case could become a landmark precedent defining where platform liability boundaries lie for AI-generated content.
Google has long taken the position that "search results are merely an intermediary for information," but for AI-generated summaries like AI Overview, a liability structure different from traditional search results is likely to apply. When AI-generated information causes actual harm, who bears responsibility? — For the first time, this question is being seriously answered in a court of law.
As the title "Truth or Consequences" suggests, in the era of AI search, accuracy is no longer merely a technical goal — it is becoming a legal obligation.
NewCore Raises $66 Million — Giving AI Agents a "Form of ID"
As AI agents begin working as "employees" inside companies, the next security challenge lies in managing not humans, but agents — so argues NewCore, which has raised $66 million in funding. TechCrunch has the story.
Once agents autonomously access data and make decisions, the traditional question of "who logged in" is no longer sufficient. Which agent accessed which data, under whose authority? There is a need for mechanisms to track and audit whether that access was appropriate.
What NewCore is proposing is an identity management platform for AI agents. Just as human identity management (IdM) became the foundation of enterprise IT, agent IdM may well become must-have infrastructure in the near future. The day when agents outnumber employees may not be far off.
Japanese Society for Artificial Intelligence Declares "AI Does Not Replace Humans" — Four Recommendations for Social Implementation
The Japanese Society for Artificial Intelligence, marking its 40th anniversary, has announced four recommendations for the social implementation of AI in Japan. ITmedia reports.
The most noteworthy element is its fundamental stance that "AI does not replace humans." While the simplistic argument that advancing AI capabilities "takes jobs away from humans" remains persistent, the society has clearly positioned itself around "human-AI collaboration." The recommendations also touch on national security and copyright, giving them the character of policy proposals that go beyond purely technical debate.
That a society with 40 years of history has laid out a direction balancing technological progress with social harmony offers a meaningful roadmap for Japan's AI policy discussions.
100B–120B-Class Models Are Disappearing from the Market — Why?
An interesting observation was made in Reddit's LocalLLaMA community. The 100B–120B-class models that emerged after GPT-OSS-120B — GLM-4.5-Air, Nemotron-3-Super, Qwen3.5-122B, and Mistral-Small-4-119B — have all gone more than three months without an update. Meanwhile, the latest releases are increasingly polarized between 25B–35B (Gemma4, Qwen3.6) and 200B+ (Step 3.5/3.7 Flash, DeepSeek-V4-Flash, MiniMax-M3, Nemotron-3-Ultra).
Behind this "hollowing out of the middle" lie hardware constraints and shifting cost-performance dynamics. With ~30B models that run on a single GPU (80GB VRAM) reaching practical performance levels, the appeal of bundling multiple GPUs to run a 120B model has faded. Meanwhile, those pursuing maximum performance have shifted to 200B+.
The community's attention is now focused on whether mid-range models will make a comeback, or whether this polarization is here to stay. Heading into H2 2026, all eyes are on whether new models in this size class will appear.
Open (YC W24) Offers "Up to $2 Million Money-Back Guarantee" for AI Agents
Open, a Y Combinator-born company, has announced an unusual guarantee: if you are not satisfied with the results of its AI agents, it will refund the full amount up to $2 million. It is making waves on Hacker News.
To the question of "who takes the loss when AI agents fail to perform as expected," Open's answer is: "we do." It is a display of confidence — but at the same time, the flip side of the fact that AI agent reliability remains uncertain.
Whether the money-back guarantee approach will spread to other companies is unclear, but a model in which AI vendors take direct responsibility for outcomes could help lower the psychological barrier to enterprise adoption.
---