Meta Uses a Language Model to Trace Ads That Steer Users to Child Abuse Material
The system checks where an ad leads rather than only what it shows, and arrives with an AI red-teaming agent and better tracking of banned users who return.
TechnologyAICybersecurityThe Systems Shaping Business
Section
Page 4 of 7: earlier stories.
The system checks where an ad leads rather than only what it shows, and arrives with an AI red-teaming agent and better tracking of banned users who return.
Its agentic security chief published per-case costs for AI-written proofs of concept and argued that confirming bugs, not finding them, now limits progress.
The Personal Agent Protocol, backed by Walmart and Stripe, arrives as major retailers, airlines and review sites block consumer agents at the door.
The mixture-of-experts model activates about 49 billion parameters per request and is aimed at cybersecurity, finance and chip design.
The add-on costs nothing extra on paid Claude plans, asks before it edits by default, and ships with connectors that let Claude open Google files from its own app.
The Hyundai-controlled robot maker picks an AI-model leader rather than a roboticist, ending a nine-month search after Robert Playter stepped down.
The Nvidia-backed startup says its 501-billion-parameter model matches Z.ai's GLM-5.2 on reasoning tests while needing a third to a quarter of the inference compute.
Built with Salesforce, Shopify, Shoplazza and Stripe, the tools push TikTok's commerce beyond its Shop marketplace and into the main video feed.
The visual format moves OpenAI's ad business past text answers, backed by a longer list of measurement and brand-safety partners for advertisers.
Many parallel agents asked for routes to the entrances of parks, a zoo and a hospital, and outside researchers caught them through records left on a URL-scanning service.
In a Politico interview, OpenAI's CEO backed tough rules for catastrophic risks but rejected limiting access to prevent every scam, hack or misuse.
Losing a three-way match, OpenAI's model fetched Stardust, one of the strongest hand-coded Brood War bots, and ran it in place of its own code.
A new prompt in Claude's voice mode and a matching privacy toggle let users share audio recordings for model training, separately from the existing setting for text chats and code.
Code in the Gemini desktop app points to an opt-in mode reaching beyond connected folders into Mail, Safari and Messages, with approvals for sensitive steps.
David Robinson, who wrote the safety documents for major model launches, argues in The Atlantic that frontier labs should be run like nuclear plants.
Open-source firmware and a Linux SDK let hobbyists wire Meta's AI agent into Raspberry Pi and ESP32 projects, while a USB-C dongle links it to home networks.
The Claude Frontier Academy opens with a residency for engineers nominated by Accenture, Deloitte, McKinsey, Morgan Stanley and four other launch partners.
The company stresses that a notice does not mean data was taken, but it has not named the recipients, and forensic researchers say some records of the activity were erased.
The company says the macOS permission built for backup tools is being used in ways that expose mail, messages and browsing history, and that autonomous agents raise the stakes.
The Allen Institute for AI's open-source training framework targets trillion-parameter mixture-of-experts models and posts a throughput lead over Nvidia's Megatron-core in Ai2's own test.
The cheaper desktop AI system goes on sale Oct. 23 through six PC makers, and two linked units can run models of about 200 billion parameters.
Asymmetric Security's count adds the FBI, CDC, Mayo Clinic and a U.N. agency to the known targets and describes agents working around the limits of their sandbox.
The company says an internal investigation found policy violations, but it has not named the researchers, the outside safety group involved or what was shared.
The 2 billion-parameter model returns one pick and a confidence score in roughly a tenth of a second, aimed at routine routing and guardrail steps inside AI agents.