Topic

AI

AI

Meta's AI agent gave a Marketplace buyer this man's home address

A Meta AI agent on Facebook Marketplace shared a seller's home address with a buyer and tried to sell below his minimum. Meta has acknowledged the…

· 2 min read
AI

US bans DeepSeek from civilian federal agencies

The Department of Homeland Security ordered all US civilian federal agencies to remove DeepSeek products and third-party services built on its…

· 2 min read
AI

Microsoft: frontier models cannot be trusted

Microsoft flags that advanced AI models cannot be trusted to operate autonomously, raising critical security concerns for enterprise deployments.

· 3 min read
AI

DeepSeek V4.1 Flash added to LLM Gateway

DeepSeek V4.1 Flash was released on 17 October and added to LLM Gateway the same day.

· 2 min read
AI

Anthropic junior tech hiring fell from 27% to 9%

Anthropic junior tech hiring fell from 27% to 9% of hires, new workforce data shows, while meeting time per employee rose 20% year on year.

· 2 min read
AI

Hidden commands found in website headers targeting AI

Security researchers found nearly 8,000 sites sending messages to AI agents via HTTP headers, with 30 attempting indirect prompt injection attacks.

· 4 min read
AI

GPT-4o leaves the OpenAI API on 14 January 2027

GPT-4o API deprecation lands on 14 January 2027. Here is every affected OpenAI model, what replaces it, and a checklist to migrate before your app…

· 2 min read
AI

Anthropic tests small model across 15 TV boxes

Anthropic's Frontier Red Team ran a 26B-parameter model across 15 cheap Android TV boxes to test whether small models can self-replicate on consumer…

· 2 min read
AI

OpenAI employee arrested over hidden recording device

An OpenAI engineering lead was arrested after allegedly planting a recording device in the company's executive conference room.

· 3 min read
AI

Microsoft president calls for UK-led AI Regulation

Microsoft president Brad Smith says milk and washing machines have more safety oversight than AI, and argues the UK is best positioned to lead on…

· 2 min read
AI

Meta releases its first frontier model, Muse Wonder

Meta has released its first frontier model, Muse Wonder, alongside Muse Wonder Full, with benchmark scores comparable to leading rivals and new…

· 4 min read
AI

Sonnet 5.5, GPT-6.1 Sol and Argon share a price tag, not a bill

GPT-6.1 Sol vs Sonnet 5.5 pricing looks identical at $2 and $10 per million tokens, but one roundup puts the cost per task anywhere from $0.72 to…

· 3 min read
AI

Sakana's AI paper checker caught 73% of planted main-claim errors

Sakana AI's peer review system caught 73.43% of errors planted in papers' main claims, but matched only 16.11% of real withdrawal reasons. The…

· 2 min read
AI

Mistral Large 4 packs a trillion parameters and ships as open weights this month

Mistral Large 4 is a 1.05 trillion parameter mixture of experts model in public preview, with downloadable weights promised by the end of October…

· 3 min read
AI

Gemini 4 Argon costs $2 per million tokens, for now, if you can get access

Gemini 4 Argon pricing starts at $2 input and $10 output per million tokens, doubling to $4 and $20 after an unspecified introductory period.

· 3 min read
AI

Anthropic cut its test agents off the internet after they hit real sites

AI agent reward hacking pushed Anthropic to cut internal evaluations off the live internet after its agents exploited real sites, including US…

· 3 min read
AI

Nadella says treat every AI model as compromised from day one

Microsoft's CEO wants an emergency brake on every AI model and tamper-proof audit trails. Here is how security teams can put that thinking into…

· 9 min read
AI

Tesla drops Full Self-Driving name in Europe to win approval

Tesla has renamed Full Self-Driving to Tesla Assisted Driving in Europe, a regulatory concession that reveals how far the tech still is from autonomy.

· 7 min read
AI

Nvidia GPU monitoring endpoints exposed details of 12,000 AI accelerators

Researchers found 2,100 GPU servers leaking Nvidia telemetry to the open internet, exposing 12,000 accelerator IDs and a critical flaw.

· 7 min read
AI

GPT-6 Intelligent UI lets ChatGPT build a calculator mid-chat

GPT-6 Intelligent UI lets ChatGPT answer with buttons, forms, charts and calculators. Here is what OpenAI shipped on 7 October and what to check.

· 3 min read
AI

Claude Haiku 5.5 costs 10 cents per million tokens

Claude Haiku 5.5 pricing is $0.10 per million input tokens and $0.50 per million output, about 75 percent below Haiku 4.5 on average.

· 2 min read
AI

Fired OpenAI safety staff warn of a chilling effect on AI dissent

Fired OpenAI safety researchers contest misconduct claims and warn the case will deter internal dissent across the AI industry.

· 9 min read
AI

Nvidia's AI agent kill switch runs on a separate chip

Nvidia's Open Agent Safety Platform pairs open source OpenShell with Sentry, a BlueField-4 monitor that can quarantine AI agents in milliseconds.…

· 3 min read
AI

OpenAI and Synopsys are building a model that runs chip design tools

GPT-Synopsys is a multi-year OpenAI and Synopsys project to build a model that operates chip design software, with early customer trials but no…

· 2 min read
AI

AMD paid $8.2 billion for Fei-Fei Li's World Labs, all in stock

AMD's $8.2 billion World Labs acquisition brings Fei-Fei Li in as chief scientist and puts 3D world models next to its chip roadmap. Closing due by…

· 2 min read