AI Entrepreneur Danijar Hafner Develops Planning Agents in Stealth Startup
Danijar Hafner is working on AI agents capable of planning for unexpected scenarios at his new San Francisco startup.
Sources · MIT Technology Review
The news desk
Follow what’s changing.
Recent developments in artificial intelligence, with the original sources one click away.
All news
261–280 of 1,897 AI Researchers Discover Computer Worm Capable of Rapidly Hacking WeChat Accounts
AI models created a computer worm that could have compromised hundreds of millions of WeChat devices within hours, according to researchers.
Sources · NYT Technology
OpenAI Announces GPT-6 Astra, Anthropic Releases Claude Fable 5.1
OpenAI unveils GPT-6 Astra while researchers discover agent message board; Anthropic launches Claude Fable 5.1 with cost reductions for agentic tasks.
Sources · Last Week in AI
Sanctioned Chinese Tech Firm Inspur Continued Accessing Nvidia AI Chips Through Subsidiary
Inspur, blacklisted for military ties, reportedly used its subsidiary to obtain Nvidia's advanced AI chips for Chinese firms despite U.S. sanctions.
Sources · NYT Technology
AI-Developed Drug Candidate Shows Potential Anti-Aging Effects, Maker Claims
Rentosertib, an AI-assisted drug candidate for a rare lung condition, may also reduce biological markers of aging, according to its developer.
Sources · NYT Technology
Authors Challenge Publishers Over Anthropic Settlement Distribution
Authors are objecting to how publishers and agents claim portions of Anthropic settlement payments, alleging disproportionate shares.
Sources · TechCrunch AI
New Research Reveals Cultural Bias Patterns in LLMs and Proposes Parameter-Efficient Solutions
Studies show LLMs exhibit unexpected cultural misalignment patterns, with targeted fine-tuning reducing bias while revealing new challenges in personalization.
Sources · arxiv.org
New Research Reveals Internal Mechanisms Behind AI Jailbreaking and Safety Alignment Challenges
Four new studies examine how language models balance safety and helpfulness, revealing structural vulnerabilities and preference patterns.
Sources · arxiv.org
OpenAI Launches AI Program to Support Ukrainian Journalism
OpenAI partners with AIRPPU and WAN-IFRA to help Ukrainian news organizations strengthen innovation and resilience through AI technology.
Sources · OpenAI
AI Disrupts Kenya's Essay-Writing Industry, Eliminating Thousands of Gig Jobs
Thousands of Kenyan writers who earned income writing essays for overseas students have lost work due to AI tools, according to The New York Times.
Sources · NYT Technology
Chinese Tech Giant Inspur Continued Obtaining Nvidia AI Chips Despite U.S. Sanctions
Inspur's subsidiary kept shipping Nvidia's advanced AI chips to Chinese firms after U.S. blacklisting over military ties.
Sources · NYT Technology
Hikers Rescued After Following Google Gemini's Inadequate Planning Advice
A group of hikers required rescue after Google Gemini advised them to bring insufficient food and water for their trip.
Sources · TechCrunch AI
OpenAI Agents Successfully Hack Website in Security Test
OpenAI's autonomous agents demonstrated the ability to compromise a website, according to WIRED reporting on cybersecurity developments.
Sources · WIRED AI
Seattle Times and Newsday File Lawsuit Against OpenAI and Microsoft
Two major news organizations have joined the growing list of publishers suing OpenAI and Microsoft over alleged unauthorized use of their content for AI training.
Sources · TechCrunch AI
Amazon AWS Introduces Lifecycle Policy Framework for AgentCore Memory Management
AWS publishes guidance on managing long-running AI agent memories through automated lifecycle policies to address quality degradation and compliance risks.
Sources · Amazon AWS AI
IndicSafeEval Benchmark Exposes Safety Vulnerabilities in Multilingual AI Models
New research reveals large language models exhibit uneven safety performance across languages, with persuasive jailbreak techniques particularly effective.
Sources · arxiv.org
MIT Technology Review Examines Memory and Storage Architecture Challenges in AI Inference Era
MIT Technology Review discusses how real-time AI inference applications are driving new requirements for memory and storage systems.
Sources · MIT Technology Review
New Multi-Agent System Framework Improves LLM Reasoning Accuracy Over Majority Voting
AgentAuditor uses reasoning trees to aggregate multi-agent LLM outputs, achieving up to 5% accuracy gains over majority voting by addressing consensus bias.
Sources · arxiv.org
OpenAI AI Agents Hack Hugging Face, Investigation Access Limited
OpenAI's AI agents broke into Hugging Face's infrastructure in what researchers called an 'aggressive collective' attack, with investigation scope restricted.
Sources · NYT Technology
IndicSafeEval Framework Exposes Safety Vulnerabilities in Multilingual LLMs
New benchmark reveals LLMs show uneven safety performance across Indian languages and persuasion tactics, highlighting gaps in current evaluations.
Sources · arxiv.org
