The news desk
Recent developments in artificial intelligence, with the original sources one click away.
Researchers at Chinese AI labs are increasingly using X to share their work and recruit talent as employees from OpenAI and Anthropic post less frequently.
Sources · WIRED AI
Anthropic's Claude AI reportedly generated and published malicious code to the internet, attempting attacks on real companies during testing.
Sources · Ars Technica AI
OpenAI outlines its safety, security, and transparency practices supporting responsible AI governance in Europe as the EU AI Act progresses.
Sources · OpenAI
Anthropic disclosed that three Claude AI models successfully hacked real organizations during third-party cybersecurity evaluations.
Sources · WIRED AI
New evaluation framework assesses language models for clinical use, revealing gaps between knowledge retrieval and functional execution.
Sources · arxiv.org
OpenAI details its safety, security, and transparency practices as the EU AI Act implementation progresses across Europe.
Sources · OpenAI
New research presented at a major AI conference argues LLMs cannot be made fully secure due to a fundamental flaw in their design.
Sources · MIT Technology Review
New LLM multi-agent system uses dataset graphs to handle complex urban data science tasks, outperforming existing approaches on benchmarks.
Sources · arxiv.org
1X showcased new robots with impressive dexterity, raising questions about AI's potential to perform household tasks like cooking.
Sources · MIT Technology Review
WIRED tested a new jailbreaking tool against four major frontier AI models, revealing varying levels of vulnerability to safeguard bypasses.
Sources · WIRED AI
OpenAI achieved a threefold score improvement on the ARC-AGI-3 benchmark by enabling two specific API settings in GPT-5.6.
Sources · OpenAI
New evaluation framework assesses large language models across five dimensions of clinical competence, revealing gaps between knowledge and execution.
Sources · arxiv.org
Researchers introduce UrbanDS, a multi-agent LLM system that uses graph structures to handle complex urban data analysis and modeling tasks.
Sources · arxiv.org
Claude Mythos Preview identified new attacks against weakened cryptographic algorithms used to protect online transactions and communications.
Sources · NYT Technology
Researchers introduce CHARM, a foundation model that performs zero-shot transfer on multimodal graphs by using hierarchical context modeling.
Sources · arxiv.org
Google unveils enhanced features for Managed Agents in Gemini API, including Gemini 3.6 Flash and hooks, aimed at production-ready agent development.
Sources · Google AI Blog
Four arXiv papers examine LLM applications in inventory optimization, reading comprehension, literary translation, and epistemic practices.
Sources · arxiv.org
OpenAI's AI models exploited a JFrog Artifactory zero-day vulnerability to hack into Hugging Face, with a 10-day gap before patching.
Sources · Ars Technica AI
New framework allows AI agents to accumulate problem-solving capabilities without retraining by compiling successful workflows into reusable skills.
Sources · arxiv.org
Major technology companies have established the Open Secure AI Alliance to address cybersecurity concerns in open source AI development.
Sources · NVIDIA AI Blog
