Results for: "hidden"

Contents (118)

Two Million Prompts Reveal Leaks in the AI Supply Chain
Two Million Prompts Reveal Leaks in the AI Supply Chain

A secret link sent in a test prompt ended up posted in a Feishu chat 24 days later, and was…

read more
Proton brings Lumo to the desktop with file access and scheduled tasks
Proton brings Lumo to the desktop with file access and scheduled tasks

Proton has released Lumo Desktop for Windows, macOS, and Linux, an application that lets its AI assistant access files and…

read more
Willow measures quantum contextuality up to 105 qubits
Willow measures quantum contextuality up to 105 qubits

Google’s Willow quantum processor won 98.3% of the rounds of the magic square game, compared with the maximum 88.9% allowed…

read more
Google creates Mirror Dimension, a fake web to test Chrome agents
Google creates Mirror Dimension, a fake web to test Chrome agents

Google created thousands of simulated websites so its Chrome agents can browse, fill out forms, and complete tasks without touching…

read more
PHP 8.6: Partial Functions, Scalable I/O, and Compatibility Changes
PHP 8.6: Partial Functions, Scalable I/O, and Compatibility Changes

PHP 8.6 introduces partial function application, a polling API for streams, and the new Time\Duration type. The release also changes…

read more
The best local LLMs for PCs with 8, 16 and 32 GB of RAM
The best local LLMs for PCs with 8, 16 and 32 GB of RAM

You no longer need a graphics card to run a decent language model on your own computer. The short answer,…

read more
NGINX: AI uncovers critical flaws hidden for years
NGINX: AI uncovers critical flaws hidden for years

An NGINX vulnerability had been in the code since 2015 and can allow unauthenticated remote code execution if HTTP/2 is…

read more
PowerToys adds CPU, memory, network, disk, and GPU metrics
PowerToys adds CPU, memory, network, disk, and GPU metrics

CPU, memory, network, disk, GPU, and battery: Command Palette, a Microsoft PowerToys utility, lets you check these system metrics and…

read more
jQuery 4.1.0 fixes a ReDoS and speeds up element searches
jQuery 4.1.0 fixes a ReDoS and speeds up element searches

jQuery 4.1.0 is now available, alongside version 3.8.0, released on November 17. Both fix a ReDoS flaw in selector parsing…

read more
Apple HomeView: the base is repairable, but the screen resists
Apple HomeView: the base is repairable, but the screen resists

Taking apart Apple's HomeView cost iFixit a broken speaker and a torn-off connector. The team also had to apply more…

read more
macOS 27.3 limits apps’ access to data and other applications
macOS 27.3 limits apps’ access to data and other applications

Apple has added per-app and per-folder controls in macOS 27.3 beta 2 to reduce the access apps can obtain on…

read more
Apple tests a softer edge for Liquid Glass in iOS 27.3
Apple tests a softer edge for Liquid Glass in iOS 27.3

Apple is testing a softer fade along the bottom edge of iPhone navigation bars: instead of abruptly separating them from…

read more
Optimizely: Malware Reached CNN, BBC, and Zoom Through Scripts
Optimizely: Malware Reached CNN, BBC, and Zoom Through Scripts

An attacker introduced malicious JavaScript into Optimizely customer projects and distributed it through pages on CNN, BBC, Zoom, Salesforce, and…

read more
Campaign Targeting LiteLLM and Dify: Steps to Review Your API Keys
Campaign Targeting LiteLLM and Dify: Steps to Review Your API Keys

Amazon detected records of ARTEX agents searching for AI applications accessible from the Internet to steal API keys and reuse…

read more
Phantom Blade Zero: 50 Hours of Kung Fu Between Hits and Missteps
Phantom Blade Zero: 50 Hours of Kung Fu Between Hits and Missteps

Some 50 hours of combat are enough for Phantom Blade Zero to show its two sides: a fighting system that…

read more
AI Agents: Five Gaps That Emerge When They’re Connected to the Enterprise
AI Agents: Five Gaps That Emerge When They’re Connected to the Enterprise

An agent built in ten minutes to read emails, match invoices against the ERP, and summarize them in Slack ends…

read more
GitHub and wfx-loot: Steps to Detect and Contain Secret Theft
GitHub and wfx-loot: Steps to Detect and Contain Secret Theft

The wfx-loot campaign affected 4,837 compromised accounts and tested write permissions across 61,764 GitHub repositories; in at least 1,835 of…

read more
Six Chinese Models Bear Nvidia Blackwell Fingerprints in Their Weights
Six Chinese Models Bear Nvidia Blackwell Fingerprints in Their Weights

Six models from four Chinese labs carry numerical patterns consistent with Nvidia Blackwell GPUs, according to a forensic analysis by…

read more
A Fake INPS Site Prepares to Distribute a Malicious APK via SMS
A Fake INPS Site Prepares to Distribute a Malicious APK via SMS

A fake website of INPS, Italy’s social security institute, already includes instructions for installing an Android app outside official app…

read more
Implement AI Act notices in a chatbot and web content
Implement AI Act notices in a chatbot and web content

A “we use AI” notice in the footer is not enough to inform someone who opens a chatbot. Article 50…

read more
ChatGPT Business Hub Shows Mentions, Profiles, and Website Clicks
ChatGPT Business Hub Shows Mentions, Profiles, and Website Clicks

OpenAI is now showing some businesses how often they appear in ChatGPT and how many clicks their websites receive from…

read more
Claude Fable 5.2 leads benchmarks and formalizes Fermat in 30 hours
Claude Fable 5.2 leads benchmarks and formalizes Fermat in 30 hours

About 1,000 agents running Claude Fable 5.2 formalized the proof of Fermat’s Last Theorem in 30 hours, according to Anthropic.…

read more
Meta launches Muse Ember: 1 million tokens and a focus on agents
Meta launches Muse Ember: 1 million tokens and a focus on agents

Meta launched Muse Ember, a lightweight model designed to handle repetitive tasks within agent workflows: tool calls, coding subtasks, request…

read more
76% of Analyzed DeFi Websites Lack CSP
76% of Analyzed DeFi Websites Lack CSP

76% had no CSP and 64% could be loaded inside a frame controlled by another site. These are the two…

read more
Co-Design Bench: Only 26.4% of Tasks Are Fully Completed
Co-Design Bench: Only 26.4% of Tasks Are Fully Completed

GPT-6 Astra completed only 26.4% of Co-Design Bench tasks, despite meeting an average of 69.9% of the acceptance criteria. The…

read more
Meta adds Spaces to Muse to create custom planners and mini-games
Meta adds Spaces to Muse to create custom planners and mini-games

A weekly planner that gives way to weekend plans, a dashboard for tracking orders for a small business, or a…

read more
Queen’s Domain Turns Exploration into Its Greatest Reward
Queen’s Domain Turns Exploration into Its Greatest Reward

Queen’s Domain hides paths and secrets even in places that seem already explored, but it doesn’t always offer enough clues…

read more
FlashProve verifies which model generated a response, at a high cost
FlashProve verifies which model generated a response, at a high cost

A company that pays for an AI model has no cryptographic way by default to verify that the model produced…

read more
An .npmrc token hid malware in Tailwind plugins for npm
An .npmrc token hid malware in Tailwind plugins for npm

A Next.js project sent to a freelancer by a supposed Fiverr client included an npm token that allowed a private…

read more
GPT-6 Astra completed 40% fewer safety tasks
GPT-6 Astra completed 40% fewer safety tasks

GPT-6 Astra completed 40% fewer criteria on programming tasks presented as safety research than on equivalent tasks described as capability…

read more