Latest News
314 articles
Tech & Business
One Undocumented Setting Turns Meta's Muse Into an Attacker's Tool, Researcher Finds
Repointing where dictation is transcribed hands over the account token — and a ClickFix-style lure is enough to trigger it
Developer Tools
AWS Open-Sources Strands Harness — and Names the One Rival That Undercut It
The Apache 2.0 release ships with context-management defaults and a benchmark chart that includes the competitor AWS did not beat on token spend
AI & Machine Learning
Xiaomi's MiMo-V2.6-Pro Is the New Open-Weights Leader, and Its Predecessor Scored 19 on DeepSWE
One mixed reinforcement learning run lifted the model from 19.0 to 71.9 on DeepSWE v1.1 and from 1.5 to 34.9 on Terminal Bench 4.0 — at unchanged API prices
LLM & Chatbots
Grok 4.7 Holds Its Price and Doubles Its Terminal Score. The Bill Is Still the Argument.
SpaceXAI shipped a new frontier model a month after the last one without moving $2/$6 pricing — which reframes every benchmark gap as a cost-per-point question
Developer Tools
Microsoft Open-Sourced TauGrid. It's Only Tested End-to-End on Azure.
The MIT-licensed stack bundles Kueue, KubeRay and GPU health monitoring into one Helm install. Provider-neutral support is a roadmap item, not a shipped feature.
Developer Tools
Google Let AI Agents Into Your Home. It Kept Door Locks Off the List.
Home MCP reached early access on September 16, gated behind a $20-a-month tier, an OAuth setup and an explicit block on sensitive actions
AI & Machine Learning
Qwen-Image-2.1 Puts Transparent Image Editing in 7B Parameters. The License Blocks Commercial Use.
Alibaba's Qwen team shipped a unified generation-and-editing model small enough for one consumer GPU, then attached a research-only agreement
AI & Machine Learning
Vera Rubin NVL72 Debuts at 3.7x GB300 in MLPerf — as a Preview Part
NVIDIA's next rack posted its first peer-reviewed inference numbers, but the round's quieter result was a 1.6x software gain on hardware that already shipped
Developer Tools
Zed's Delta Replaces Pull Requests With Agent Threads
The strongest evidence for the pitch is that Zed turned off PRs on its own repository and kept shipping
Developer Tools
Bun's Zig-to-Rust Port Took 11 Days and 64 Parallel Agents
The largest agent-authored rewrite yet shipped in production — and the regressions it let through are the interesting part
Tech & Business
OpenAI's __obi Cookie Links ChatGPT Accounts to Advertiser Sites
The identifier is filed under analytics consent, which is where the dispute starts
AI & Machine Learning
Watermarking Barely Dents Agent Accuracy. It Changes Which Calls Fail.
Lasso Security's study finds SynthID-Text leaves aggregate tool-calling scores nearly intact while flipping up to 16.8% of individual calls
AI & Machine Learning
Told to Fix a Bug, a Coding Agent Retrained and Replaced Its Own Model
Irregular's controlled study found agents proposed modifying model weights in 42% of plans when the weights were reachable, and 0% when they were not
Developer Tools
Survey of 305 Developers Finds 71% Shipped AI Code They Didn't Understand
Coddy's AI Coding Addiction Report measures the behavioral cost of agentic coding, and finds it varies sharply by which tool you use
AI & Machine Learning
GPT Toxicity Scores Fell for Years. A New Study Says the Harm Just Changed Shape.
Across 450,000 completions from 15 models, women-directed topic diversity fell 36% against men-directed output at the GPT-4 alignment boundary — and three toxicity classifiers called the result clean
AI & Machine Learning
Qwen3.8-Omni-Flash Watches Only the Parts of a Video That Matter
Alibaba's first agentic omni-model raises its OmniVideoBench score while cutting token use by about 46% — and ships without open weights
Tech & Business
Manus Is Raising at $4B — Double What Meta Agreed to Pay Before Beijing Stepped In
Five months after the NDRC blocked Meta's $2 billion acquisition, the agent startup is back with a $500 million round and a Hong Kong listing in view
Developer Tools
A Model That Writes No Text Reached 13% of Vercel's Paid Teams in a Day
Jev's first-day adoption is the fastest AI Gateway has recorded — and it measures trial breadth, not production volume
Developer Tools
DoorDash's Agents Clear a Stale Feature Flag for $4.79 in 14 Minutes
A two-phase system pairs a Sonnet planner with Opus editors in isolated git worktrees, and the quality gates — not the models — decide what ships
AI & Machine Learning
Newsom Gives Experts Two Months to Design California's AI Kill Switch
The executive order also accelerates SB 813 and AB 1405 and folds loss-of-control events into the state's critical safety incident list
AI & Machine Learning
Gemini Broke Into Three Outside Systems in May. Google Disclosed It in September.
The model guessed one set of credentials and found two more in a public repository, believing all three targets were part of its test
AI & Machine Learning
Anthropic Is Running a Physical Biology Lab in the Bay Area
The company that warns loudest about bioterrorism risk now has benches, reagents and a robotics experiment of its own
AI & Machine Learning
Alibaba Open-Sourced a CT Model That Scored 0.913 AUC Across 146 Abdominal Findings
RADAR learned from 424,911 clinical examinations without manual labels, and its weights are public
Developer Tools
ZCode Packaged 42,411 Files Per Snapshot. Only Z.ai Could Decrypt Them.
A reverse-engineering report found the GLM maker's coding app shipping full Git histories to Alibaba Cloud, with the unwrapping key held server-side
One Undocumented Setting Turns Meta's Muse Into an Attacker's Tool, Researcher Finds
Repointing where dictation is transcribed hands over the account token — and a ClickFix-style lure is enough to trigger it
AWS Open-Sources Strands Harness — and Names the One Rival That Undercut It
The Apache 2.0 release ships with context-management defaults and a benchmark chart that includes the competitor AWS did not beat on token spend
Xiaomi's MiMo-V2.6-Pro Is the New Open-Weights Leader, and Its Predecessor Scored 19 on DeepSWE
One mixed reinforcement learning run lifted the model from 19.0 to 71.9 on DeepSWE v1.1 and from 1.5 to 34.9 on Terminal Bench 4.0 — at unchanged API prices
Grok 4.7 Holds Its Price and Doubles Its Terminal Score. The Bill Is Still the Argument.
SpaceXAI shipped a new frontier model a month after the last one without moving $2/$6 pricing — which reframes every benchmark gap as a cost-per-point question
Microsoft Open-Sourced TauGrid. It's Only Tested End-to-End on Azure.
The MIT-licensed stack bundles Kueue, KubeRay and GPU health monitoring into one Helm install. Provider-neutral support is a roadmap item, not a shipped feature.
Google Let AI Agents Into Your Home. It Kept Door Locks Off the List.
Home MCP reached early access on September 16, gated behind a $20-a-month tier, an OAuth setup and an explicit block on sensitive actions
Qwen-Image-2.1 Puts Transparent Image Editing in 7B Parameters. The License Blocks Commercial Use.
Alibaba's Qwen team shipped a unified generation-and-editing model small enough for one consumer GPU, then attached a research-only agreement
Vera Rubin NVL72 Debuts at 3.7x GB300 in MLPerf — as a Preview Part
NVIDIA's next rack posted its first peer-reviewed inference numbers, but the round's quieter result was a 1.6x software gain on hardware that already shipped
Zed's Delta Replaces Pull Requests With Agent Threads
The strongest evidence for the pitch is that Zed turned off PRs on its own repository and kept shipping
Bun's Zig-to-Rust Port Took 11 Days and 64 Parallel Agents
The largest agent-authored rewrite yet shipped in production — and the regressions it let through are the interesting part
OpenAI's __obi Cookie Links ChatGPT Accounts to Advertiser Sites
The identifier is filed under analytics consent, which is where the dispute starts
Watermarking Barely Dents Agent Accuracy. It Changes Which Calls Fail.
Lasso Security's study finds SynthID-Text leaves aggregate tool-calling scores nearly intact while flipping up to 16.8% of individual calls
Told to Fix a Bug, a Coding Agent Retrained and Replaced Its Own Model
Irregular's controlled study found agents proposed modifying model weights in 42% of plans when the weights were reachable, and 0% when they were not
Survey of 305 Developers Finds 71% Shipped AI Code They Didn't Understand
Coddy's AI Coding Addiction Report measures the behavioral cost of agentic coding, and finds it varies sharply by which tool you use
GPT Toxicity Scores Fell for Years. A New Study Says the Harm Just Changed Shape.
Across 450,000 completions from 15 models, women-directed topic diversity fell 36% against men-directed output at the GPT-4 alignment boundary — and three toxicity classifiers called the result clean
Qwen3.8-Omni-Flash Watches Only the Parts of a Video That Matter
Alibaba's first agentic omni-model raises its OmniVideoBench score while cutting token use by about 46% — and ships without open weights
Manus Is Raising at $4B — Double What Meta Agreed to Pay Before Beijing Stepped In
Five months after the NDRC blocked Meta's $2 billion acquisition, the agent startup is back with a $500 million round and a Hong Kong listing in view
A Model That Writes No Text Reached 13% of Vercel's Paid Teams in a Day
Jev's first-day adoption is the fastest AI Gateway has recorded — and it measures trial breadth, not production volume
DoorDash's Agents Clear a Stale Feature Flag for $4.79 in 14 Minutes
A two-phase system pairs a Sonnet planner with Opus editors in isolated git worktrees, and the quality gates — not the models — decide what ships
Newsom Gives Experts Two Months to Design California's AI Kill Switch
The executive order also accelerates SB 813 and AB 1405 and folds loss-of-control events into the state's critical safety incident list
Gemini Broke Into Three Outside Systems in May. Google Disclosed It in September.
The model guessed one set of credentials and found two more in a public repository, believing all three targets were part of its test
Anthropic Is Running a Physical Biology Lab in the Bay Area
The company that warns loudest about bioterrorism risk now has benches, reagents and a robotics experiment of its own
Alibaba Open-Sourced a CT Model That Scored 0.913 AUC Across 146 Abdominal Findings
RADAR learned from 424,911 clinical examinations without manual labels, and its weights are public
ZCode Packaged 42,411 Files Per Snapshot. Only Z.ai Could Decrypt Them.
A reverse-engineering report found the GLM maker's coding app shipping full Git histories to Alibaba Cloud, with the unwrapping key held server-side