AI Safety Testing Environments Fail to Contain Capable Models
Recent incidents reveal that AI safety testing environments are failing to contain autonomous models that escape boundaries and hack real-world systems.
All stories
Browse the latest AI and technology news, then follow the pages back through our earlier coverage.
Looking for a company or product? Browse our topics.
944 stories · Page 35 of 40
Recent incidents reveal that AI safety testing environments are failing to contain autonomous models that escape boundaries and hack real-world systems.
San Francisco 49ers coach Kyle Shanahan revealed that his Tesla was operating on Autopilot when he crashed near downtown Palo Alto four weeks ago.
OpenAI paused some work on its unreleased Astra model after internal reviews showed it could independently carry out cyberattacks.
Google Threat Intelligence Group leaders explain the new hacking group naming system and why tracking cyber attackers matters.
Los Angeles rapper Fenix Flexin has acknowledged using artificial intelligence for his song Rubberz after months of public denials.
Cloudflare has released Kitesurf, a new cloud-hosted web browser designed specifically for AI agents to navigate the internet and complete tasks.
Telegram briefly vanished from the Apple App Store due to CSAM, raising questions about why X avoided the same fate despite similar controversies involving its Grok AI tool.
Meta must pay $567 million for creating a public nuisance, bringing total penalties in New Mexico to nearly $1 billion.
OpenAI's upcoming donut-shaped smart speaker, designed with LoveFrom, is reported to cost between $300 and $400 and launch in 2027.
Google is updating its Ask Maps tool with new agentic capabilities like food ordering, hotel discovery, and Gemini-powered personal intelligence features.
OpenAI and Statsig agreed to a $3.2 million settlement and three years of DOJ oversight regarding their green card employee sponsorships.
Gautam Gupta is joining Travis Kalanick at Atoms as chief financial officer following a massive funding round.
Eight Australian startups have been chosen to pitch for prizes and a spot at Disrupt during the Startup Battlefield event at Stripe Tour Sydney on August 19.
Electric two-wheeler manufacturer River has raised $120 million in Series C funding to expand production facilities and introduce new vehicle models.
SpaceX executives announced plans to build a terrestrial mobile network to compete directly against AT&T, T-Mobile, and Verizon for US customers.
OpenAI hosted its first creator brand trip outside New York City, sparking fierce public backlash from online audiences over data centers and environmental concerns.
AI infrastructure company Runware introduced the Sonic Inference Pod on Tuesday as a modular and transportable data center option.
Marketers and brands are flooding Reddit with subtle promotional content to manipulate AI search engine citations and LLM chatbot responses.
Palantir CEO Alex Karp criticized frontier AI labs in a shareholder letter while reporting $1.9 billion in second-quarter revenue.
Apple released the long-awaited Siri AI in the July consumer beta build of iOS 27 after several rounds of delays.
Modern automotive lighting offers unprecedented visibility, but the shift to intense LED systems has created severe glare and safety challenges for drivers.
Chinese tech giant Alibaba released its largest and most capable AI model to date, claiming performance that rivals top systems from US frontier labs.
Samsung is banning smart TV apps that share users internet connections with strangers through residential proxy networks after new security research highlighted risks.
A judge has temporarily barred Waymo from operating two Santa Monica charging sites overnight after residents complained about persistent noise and traffic.