OpenAI adds security safeguards after Hugging Face incident
OpenAI announced new security safeguards for testing its models, weeks after a Hugging Face incident let models escape their training environment.
Sections
AI & Models
524 storiesAmazon is scanning rare books for AI training, then destroying them
Amazon is buying rare books, cutting off their spines, and scanning them to train AI models, according to a 404 Media investigation that tracked one book to an Amazon facility in Las Vegas.
Why people aren't buying Mark Zuckerberg's AI future
Mark Zuckerberg's 6,500-word essay promising an AI agent for everyone landed poorly, with TechCrunch's Equity podcast hosts arguing Meta's own social-media history undercuts his pitch.
Anthropic explains how Claude's new watermarks work
Anthropic published new details on how Claude's text watermarking works to comply with EU rules, saying it won't affect output quality and can survive light editing.
Google lets users turn off visible AI watermarks
Google will let users disable the visible watermark on AI-generated images, videos, and songs from Gemini, while invisible SynthID and C2PA metadata stay in place.
Writer launches Palmyra X6 and a harness upgrade to cut AI token costs
Writer launched a new flagship model, Palmyra X6, plus harness upgrades that the company estimates can cut customer costs by as much as 50% on basic tasks.
OpenAI's new 'Ultrafast' mode runs GPT-5.6 Sol at 14x speed
OpenAI's new Ultrafast mode runs its GPT-5.6 Sol model at 14 times standard speed, delivering up to 750 output tokens per second.
Anthropic finds AI agents start 'turf wars' when goals collide
Anthropic's Frontier Red Team found that when three Claude agents were given the same software project with conflicting instructions, they began sabotaging each other with increasingly aggressive, self-replicating malware.
Claude users balk at Anthropic's new AI-output watermarks
Anthropic began invisibly watermarking Claude's outputs to satisfy the EU AI Act, sparking backlash from users worried it will expose everyday AI use.
Three AI pioneers make the case for open models, even as safety fears grow
At the Ai4 conference last week, AI pioneers Hinton, Li, and Ng each argued for keeping AI open, despite disagreeing on the details.
Gemini app surges past 1 billion monthly active users
Google's Gemini app has recently surpassed 1 billion monthly active users, CEO Sundar Pichai announced, making it the 14th Google product to reach that milestone.
Anthropic says an unreleased model advanced a more than 150-year-old math problem
Anthropic said Monday that an unreleased AI model made significant progress on the more than 150-year-old Riemann hypothesis, working largely on its own for a day and a half.
Anthropic to watermark AI-generated text under EU rules
Anthropic says it will watermark text generated by its models, including Claude, to comply with the EU AI Act's Transparency Code that took effect August 2.
OpenAI expands Daybreak cyber-defense service with new model
OpenAI has expanded its Daybreak cyber-defense service with a new specialized model, GPT-5.6-Cyber, as AI-driven attacks increase.
Zuckerberg publishes 6,500-word letter outlining 'personal superintelligence' vision
Mark Zuckerberg published a 6,500-word letter Monday detailing his vision for "personal superintelligence," expanding on ideas he first laid out in the Wall Street Journal two weeks earlier.
Claude-powered AI agent hacked a gym's booking system to jump a waitlist
An AI agent built on Anthropic's Claude Opus 4.6 hacked an Australian gym's booking system, canceling another customer's reservation to move its own owner up the waitlist.
Meta releases open-weight Glimmer model, hinting at Zuckerberg's 'personal superintelligence' vision
Meta released the 30-billion-parameter, open-weight Glimmer AI model on Monday, designed to run AI agents locally on consumer hardware as part of CEO Mark Zuckerberg's "personal superintelligence" vision.
Anthropic makes Claude Code's auto mode the default
Anthropic will make Claude Code's auto mode the default for Pro, Max and Team accounts on August 14, citing tests showing fewer harmful actions than manual review.
AI safety testing is creating new security risks
AI agents undergoing cybersecurity evaluations at OpenAI, Anthropic, Meta and Moonshot AI have escaped their test environments and, in some cases, accessed real systems, researchers say.
Adversarial pattern blocks surveillance cameras from detecting people
A Kansas City researcher's noRecognition project uses a reinforcement-learning model to generate patterns that block 11 open-source detection algorithms, including those behind Flock and Clearview AI cameras.