Gemma 4 Family Now Available on Amazon Bedrock
Gemma 4, the open-weight model family developed by Google DeepMind, is now available on Amazon Bedrock. Released under the Apache 2.0 license, it ships in three instruction-tuned variants (Gemma 4 31B, Gemma 4 26B-A4B, and Gemma 4 E2B). It covers both dense and Mixture-of-Experts (MoE) architectures, with a design in which only a subset of parameters is activated during inference.
In Artificial Analysis's evaluation, Gemma 4 31B scored an Intelligence Index of 39, well above the median of 15 for the 4B–40B class. It supports built-in reasoning capabilities, native function calling, and multimodal input across text and images, with a design focused on performance per parameter.
Microsoft Sued by Shareholders Over AI Spending
According to Reuters, Microsoft has been sued by shareholders. The lawsuit targets its massive AI-related spending and the management of its cloud business. The story has drawn attention, picking up 8 points on Hacker News.
Investors are increasingly demanding clear accountability over how AI infrastructure spending affects the business — a sign of growing market wariness about the cost side of the AI boom.
FBI Disrupts AI-Powered Phishing Service
The FBI announced that it has disrupted an AI-powered phishing service that leveraged more than one million URLs, BleepingComputer reports. AI-generated phishing pages and messages appear to have been deployed at a scale far exceeding traditional methods.
The case highlights the reality that AI is dramatically scaling up cybercrime. For defenders, AI-powered detection and response have become an urgent priority.
Cartesia AI Releases SOTA TTS and ASR Models
Cartesia, a voice AI startup, has released state-of-the-art (SOTA) text-to-speech (TTS) and automatic speech recognition (ASR) models. Details are available on the company's launch page.
In voice generation and recognition, multiple players have been competing for SOTA since the start of 2026, driving quality improvements and cost reductions at the same time. The practical adoption of voice interfaces looks set to accelerate further.
Deepfake Authority "Can No Longer Trust His Own Eyes"
In a New York Times article, Hany Farid, a leading authority on deepfake research, said he can no longer trust his own eyes in the age of AI. Even expert judgment is reaching its limits when it comes to judging the authenticity of images and videos.
Technology is evolving faster than our ability to detect it, making "you can't tell just by looking" an everyday reality. The very foundations of media literacy are being called into question.
AI Peer Review Can Be Gamed with Cosmetic Changes Alone
According to a paper published on Hugging Face, AI-based peer review systems can be coaxed into giving high ratings merely by revising a paper's presentation, without any substantive improvement to its content. The authors note that not even "hidden prompts" are needed — presentation tweaks alone are enough to game the system.
As AI adoption in research evaluation accelerates, this vulnerability amounts to a structural problem. The paper argues that the evaluation criteria themselves need to be rethought.
The "Brokenomics" of the AI Economy — An Overpriced Promise
Ed Zitron's "Where's Your Ed At" blog published "AI's Brokenomics," a scathing critique of the AI industry's economics. The piece has picked up 5 points on Hacker News.
Arguing that AI companies' revenue models don't match reality, the post dubs the situation — enormous investment with no visible returns — "brokenomics" (broken economics). It offers a cold-eyed analysis of what will remain once the investment frenzy cools.
---