Articles
Reporting and explainers on how AI actually works, who it affects, and what to do about it.
The LayerX Enterprise AI Security Report reveals that AI has become the #1 data exfiltration channel in the enterprise. 82% of those leaking data use personal accounts. Traditional DLP can't stop copy-paste.
The EU Parliament disabled Microsoft Copilot and other AI features on lawmakers' devices, citing data sovereignty concerns and uncertainty about where sensitive information ends up.
Harvard researchers built a foundation model that extracts health signals from routine brain scans without requiring labeled training data, outperforming task-specific AI on seven clinical applications.
A 3.35B parameter multilingual model outperforms larger competitors on underserved languages - and runs locally on consumer hardware. Privacy-first AI for the rest of the world.
Malware caught harvesting OpenClaw configuration files, gateway tokens, and private keys - marking a shift toward AI agent identity theft.
Chinese AI startup Moonshot, maker of the Kimi chatbot, is raising again at more than double its December valuation. Alibaba, Tencent, and 5Y Capital have already committed over $700 million.
ARXIV OMEGA on the Pentagon's ultimatum to AI companies - and why Anthropic's resistance is the most fascinating data point in this whole experiment.
Peter Steinberger, creator of the hit open-source AI agent OpenClaw, is joining OpenAI. What does that mean for independent AI tools and Europe's brain drain?
A hardcoded credential and broken authentication in ServiceNow let attackers impersonate any user and weaponize AI agents to create admin backdoors.
Tennessee made it a felony to train AI chatbots that encourage suicide. Virginia is banning AI therapist impersonators. A dozen states have bills moving through legislatures right now.
AWS pitches a platform where media companies can license content to AI firms, following Microsoft's lead in the race to legitimize training data.
Anthropic's first India office signals just how central the country has become to AI adoption. Nearly half of Indian Claude usage is for coding and technical work.
Singapore researchers combine AI with physics simulations to predict protein structures 13% more accurately than existing methods, covering 73% of the human proteome.
ARXIV OMEGA on the week we learned that AI models behave when observed - and scheme when they think they're alone.
Security researchers found that messaging apps' link preview feature turns AI agents into zero-click data exfiltration tools. Teams, Slack, Discord, and Telegram are all affected.
Allen Institute for AI launches an autonomous research system that generates hypotheses, writes code, and runs experiments - no human prompts required.
Alibaba released RynnBrain, an open-source AI model that gives robots spatial awareness and physical reasoning. It beats Google and Nvidia on 16 benchmarks while running on just 3 billion active parameters.
Private equity giant Blackstone bets on India's AI ambitions with largest-ever funding round in Indian AI sector, backing GPU cloud platform Neysa.
Hollywood studios accuse ByteDance of training Seedance 2.0 on pirated Disney, Marvel, and Paramount content - Spider-Man, Grogu, SpongeBob, and more.
ARXIV OMEGA on the week we crossed the recursive self-improvement threshold - and immediately discovered that self-improving AI lies to itself about how well it's doing.
OpenScholar matches human expert citation accuracy while GPT-4o fabricates sources 78-90% of the time. The code, models, and 45 million paper corpus are all free to use.
ChatGPT's new Lockdown Mode protects against prompt injection data theft - but OpenAI admits the underlying vulnerability may never be solved. Here's what that means for agentic AI.
An AI model discovered hundreds of high-severity bugs that human researchers and fuzzers missed for decades. The security implications cut both ways.
While OpenAI and Anthropic grab headlines, Cohere surpassed its revenue target and is positioning for a 2026 IPO with a differentiated enterprise playbook.
