Meta unveils Muse Image and Muse Video, its first in-house media generators
Meta Superintelligence Labs previewed Muse Image and Muse Video, its first homegrown media-generation models, and built 'agentic tool use' directly into the image model instead of treating generation as passive prompting.
Superhuman acquires AI-detector GPTZero to build a persistent 'authenticity layer'
Email company Superhuman is acquiring GPTZero, the AI-text detector that grew to 19 million users, and plans to turn one-off detection into an always-on authenticity layer that travels with people as they read and write.
Authors file a new $75M suit against Anthropic as scholars redefine what an AI 'copy' is
Authors who opted out of Anthropic's $1.5 billion Bartz settlement filed a separate $75 million copyright suit over how their books were sourced, landing the same week two law scholars argued that AI weights may be 'probabilistic copies' the law hasn't defined yet.
Flathub bans AI 'slop' apps - and data shows 73% of them were already abandoned
After Linux app store Flathub moved to ban AI-generated 'slop' applications, a developer tracked 120 flagged repositories and found 88 of them - 73% - had been abandoned or deleted from GitHub within months, turning an anecdote into a measured fact.
A consultancy is charging $10,000 a week to delete AI-generated code
A boutique agency is openly marketing a service to remove AI-generated 'slop' from codebases at $10,000 a week, and a front-page Hacker News debate showed developers increasingly see reviewing incoherent AI pull requests as a real, expensive liability.
A startup router is giving away 100 million tokens of Kimi, MiniMax and GLM
API aggregator Dahl Inference is handing out 100 million free tokens across top open-weight Chinese models like Kimi K2.6 and MiniMax M2.7 - not a price cut from the labs themselves, but a router burning money to win users amid a glut of cheap compute.
A red-teaming study cracked production AI agents 94% of the time
A new framework called Vera stress-tested real AI agent systems like Claude Code and Hermes in sandboxes and found that multi-channel attacks succeeded 93.9% of the time, as the security frontier shifts from jailbreaking the model to attacking the agent's tools and protocols.
A giant benchmark tested 24 optimizers - and AdamW's edge held up
OmniOpt ran a controlled bake-off of more than two dozen modern training optimizers across model sizes from 60M to 1B parameters, and its main lesson is deflating: no challenger cleanly dethrones AdamW, because an optimizer's advantage depends heavily on scale, task, and tuning budget.
'World model' was too vague, so researchers wrote a 58-page definition
With 'world model' stretched to mean everything from video generators to robot simulators, a 58-page technical report set out to pin down a single scientific definition - as a wave of new papers pushed the field from world-models-as-pretty-video toward world-models-as-working-tools.
New methods page an AI's memory in and out like an operating system
Two new techniques treat a language model's long-context memory like an operating system's memory hierarchy - keeping coarse summaries on the GPU and paging compressed detail out to the CPU - with one, SeKV, cutting GPU memory use by 53% at 128,000 tokens while avoiding the errors that plague simpler methods.