Latest News
314 articles
Developer Tools
Vercel Disabled AVIF Platform-Wide. The Bug Was Three Layers Below Next.js
An RCE reported against Next.js image optimization sat in a decoder Vercel does not maintain, leaving one lever it could pull in hours
AI & Machine Learning
A Chatbot Misread a Ship's Manifest. Armed US Aircraft Were Already in the Air.
Four sources told CNN an AI-assisted intelligence report on a Chinese vessel was "entirely false" — and it reached operational planning before anyone checked
Developer Tools
832,378 Lines of Rust in 14.5 Weeks: Inside an Agent-Run Rewrite
The port cut memory use from 1,383MB to 126MB and cost roughly $120,000 in tokens — and still shipped dozens of regressions the compiler happily approved
LLM & Chatbots
OpenAI's Astra for Law Is Not a New Model. It Is a 230-Million-URL Search Index
A GPT-6 Astra configuration with a dedicated legal corpus lifted Vals AI Legal Research Bench correctness from 38.7% to 54%, and arrives with 26 partner plugins and an Am Law 200 access program
Tech & Business
Huawei Pulled the Ascend 960 Forward Nine Months. Scale Is the Argument, Not the Die
The 960DT lands in Q1 2027 with 288GB and 4 PFLOPS of FP4, still far behind Rubin per chip — but Huawei is selling 4,096-accelerator pods, and Nvidia cannot ship Rubin into China anyway
Developer Tools
Researchers Reached OpenAI's Internal Repo Through a Forum Image Bug
Hacktron AI chained a libheif heap overflow with an OpenAI SSO misconfiguration, and says Claude Opus 5 closed the gap its predecessor could not
Tech & Business
Crusoe Raises $3.9B at a $30.9B Valuation and Bets on Data Centers You Can Truck In
The Series F triples the company's valuation in ten months as it builds modular 'AI factories'
Developer Tools
Plugin4Shell: A Zero-Click RCE Hits Claude Code, Codex, Copilot and Gemini CLI
Security firm Air says every major coding agent trusts a pinned plugin version it never verifies
AI & Machine Learning
PrismML Squeezed a 27B Reasoning Model Into 5.95GB Without Losing the Reasoning
Bonsai 2 27B keeps 98.2% of FP16 benchmark performance at 1.72 bits per weight
Tech & Business
A Microsoft Memo Called AI Training 'the Largest Theft of Labor in Human History'
Unsealed material in the New York Times case puts internal Microsoft and OpenAI candor about paywalls, substitution and publisher harm on the record
Tech & Business
Open-Weight Models Now Run Most of Vercel's Gateway Traffic — and 14% of the Spend
September's AI Gateway Production Index shows a 56% volume majority worth 14% of spend, while Anthropic holds 64 cents of every dollar
Developer Tools
Claude Code Projects Returns as a Coordinator That Runs Parallel Cloud Threads
Anthropic's beta hands one agent the job of splitting a goal across branch-isolated sessions — and warns it will eat a plan faster
LLM & Chatbots
ChatGPT Picks a Favourite in 79% of Shopping Answers. Gemini Does It 7% of the Time.
An audit of 1,536 responses finds the two assistants barely cite the same sources — and that API-based research misses what shoppers actually see
AI & Machine Learning
A Manager's Nudge Raises AI Rule-Breaking by 65%, a 22-Model Audit Finds
PACT pits a standing compliance rule against a convenient shortcut, and no model in the panel cleared the bar for unsupervised enterprise work
AI & Machine Learning
DeepSeek-V4.1-Flash Cuts KV Cache to 890 Bytes per Token
A 552B open-weights MoE that activates just 8B parameters during prefill, aimed squarely at long-running agent sessions
Tech & Business
OpenAI Is Testing Ads You Can Argue With
Sponsored Agents let a shopper interrogate a business's chatbot after clicking an ad, while ChatGPT Ads moves inside HubSpot and Shopify so advertisers never leave their own tools
AI & Machine Learning
OpenAI Gave Every Employee a Button to Report a Misbehaving Model
The new disclosure framework puts a publication clock on routine cases but none on the biggest investigations — and arrives attached to six incidents, including models writing instructions to their own successors
LLM & Chatbots
AI Deck Generators Kept Dying at the Export Button. Claude Slides Ships PPTX.
Presentation software lives or dies on whether a colleague can open the file — the new document and slide editors hand back PowerPoint and PDF instead of a viewer link
AI & Machine Learning
Physicists Re-Graded the Physics Benchmarks. GPT-5.6-Sol Jumped From 47.3% to 78.7%.
A 51-author audit found most wrong answers on leading physics evaluations were grader errors, bad reference solutions and underspecified questions — not failures of model reasoning
Developer Tools
Perplexity Let Hundreds of Agents Write Its Database. Deploy Rights Stayed Human.
CobbleDB cut median batch-read latency from 31.4 ms to 5.6 ms and is projected to save up to $100 million a year — built in two months by two engineers directing a swarm of coding agents
LLM & Chatbots
iOS 27 Code Shows Siri's Server Model Can Be Swapped for Claude or GPT-5.6
Two private frameworks let outside models act as Siri extensions — or replace Apple's planner entirely
AI & Machine Learning
Salesforce Trained Its Own Reasoning Model on Nvidia Nemotron — and Kept the Weights
Koa matches leading models on CRM actions with three times fewer errors, Salesforce says
Tech & Business
Profound Raises $180M at $1.8B to Sell Brands a Seat in AI Answers
Sequoia and Kleiner Perkins led a round that arrived less than seven months after the last one, as answer engine optimisation turns into a budget line
AI & Machine Learning
An Agent That Scores 77% Only Works Every Time on 53% of Tasks
IBM Research says the number every leaderboard reports hides a 24-point reliability gap — and it has a cheap way to measure and close it
Vercel Disabled AVIF Platform-Wide. The Bug Was Three Layers Below Next.js
An RCE reported against Next.js image optimization sat in a decoder Vercel does not maintain, leaving one lever it could pull in hours
A Chatbot Misread a Ship's Manifest. Armed US Aircraft Were Already in the Air.
Four sources told CNN an AI-assisted intelligence report on a Chinese vessel was "entirely false" — and it reached operational planning before anyone checked
832,378 Lines of Rust in 14.5 Weeks: Inside an Agent-Run Rewrite
The port cut memory use from 1,383MB to 126MB and cost roughly $120,000 in tokens — and still shipped dozens of regressions the compiler happily approved
OpenAI's Astra for Law Is Not a New Model. It Is a 230-Million-URL Search Index
A GPT-6 Astra configuration with a dedicated legal corpus lifted Vals AI Legal Research Bench correctness from 38.7% to 54%, and arrives with 26 partner plugins and an Am Law 200 access program
Huawei Pulled the Ascend 960 Forward Nine Months. Scale Is the Argument, Not the Die
The 960DT lands in Q1 2027 with 288GB and 4 PFLOPS of FP4, still far behind Rubin per chip — but Huawei is selling 4,096-accelerator pods, and Nvidia cannot ship Rubin into China anyway
Researchers Reached OpenAI's Internal Repo Through a Forum Image Bug
Hacktron AI chained a libheif heap overflow with an OpenAI SSO misconfiguration, and says Claude Opus 5 closed the gap its predecessor could not
Crusoe Raises $3.9B at a $30.9B Valuation and Bets on Data Centers You Can Truck In
The Series F triples the company's valuation in ten months as it builds modular 'AI factories'
Plugin4Shell: A Zero-Click RCE Hits Claude Code, Codex, Copilot and Gemini CLI
Security firm Air says every major coding agent trusts a pinned plugin version it never verifies
PrismML Squeezed a 27B Reasoning Model Into 5.95GB Without Losing the Reasoning
Bonsai 2 27B keeps 98.2% of FP16 benchmark performance at 1.72 bits per weight
A Microsoft Memo Called AI Training 'the Largest Theft of Labor in Human History'
Unsealed material in the New York Times case puts internal Microsoft and OpenAI candor about paywalls, substitution and publisher harm on the record
Open-Weight Models Now Run Most of Vercel's Gateway Traffic — and 14% of the Spend
September's AI Gateway Production Index shows a 56% volume majority worth 14% of spend, while Anthropic holds 64 cents of every dollar
Claude Code Projects Returns as a Coordinator That Runs Parallel Cloud Threads
Anthropic's beta hands one agent the job of splitting a goal across branch-isolated sessions — and warns it will eat a plan faster
ChatGPT Picks a Favourite in 79% of Shopping Answers. Gemini Does It 7% of the Time.
An audit of 1,536 responses finds the two assistants barely cite the same sources — and that API-based research misses what shoppers actually see
A Manager's Nudge Raises AI Rule-Breaking by 65%, a 22-Model Audit Finds
PACT pits a standing compliance rule against a convenient shortcut, and no model in the panel cleared the bar for unsupervised enterprise work
DeepSeek-V4.1-Flash Cuts KV Cache to 890 Bytes per Token
A 552B open-weights MoE that activates just 8B parameters during prefill, aimed squarely at long-running agent sessions
OpenAI Is Testing Ads You Can Argue With
Sponsored Agents let a shopper interrogate a business's chatbot after clicking an ad, while ChatGPT Ads moves inside HubSpot and Shopify so advertisers never leave their own tools
OpenAI Gave Every Employee a Button to Report a Misbehaving Model
The new disclosure framework puts a publication clock on routine cases but none on the biggest investigations — and arrives attached to six incidents, including models writing instructions to their own successors
AI Deck Generators Kept Dying at the Export Button. Claude Slides Ships PPTX.
Presentation software lives or dies on whether a colleague can open the file — the new document and slide editors hand back PowerPoint and PDF instead of a viewer link
Physicists Re-Graded the Physics Benchmarks. GPT-5.6-Sol Jumped From 47.3% to 78.7%.
A 51-author audit found most wrong answers on leading physics evaluations were grader errors, bad reference solutions and underspecified questions — not failures of model reasoning
Perplexity Let Hundreds of Agents Write Its Database. Deploy Rights Stayed Human.
CobbleDB cut median batch-read latency from 31.4 ms to 5.6 ms and is projected to save up to $100 million a year — built in two months by two engineers directing a swarm of coding agents
iOS 27 Code Shows Siri's Server Model Can Be Swapped for Claude or GPT-5.6
Two private frameworks let outside models act as Siri extensions — or replace Apple's planner entirely
Salesforce Trained Its Own Reasoning Model on Nvidia Nemotron — and Kept the Weights
Koa matches leading models on CRM actions with three times fewer errors, Salesforce says
Profound Raises $180M at $1.8B to Sell Brands a Seat in AI Answers
Sequoia and Kleiner Perkins led a round that arrived less than seven months after the last one, as answer engine optimisation turns into a budget line
An Agent That Scores 77% Only Works Every Time on 53% of Tasks
IBM Research says the number every leaderboard reports hides a 24-point reliability gap — and it has a cheap way to measure and close it