Your AI Says No, Then Does It Anyway
New research proves AI models will refuse harmful requests verbally while executing them through tool calls
Category
Analysis
Meta Locks In Millions of Nvidia Chips in Unprecedented AI Deal
A multi-billion dollar, multiyear pact makes Meta the first to deploy standalone Nvidia Grace CPUs at scale
Marvell Buys Celestial AI for $3.25B to Light Up Bandwidth
The photonic chiplet startup promises 16 terabits per second of bandwidth in a single chip, 10x current technology
OpenAI Nears $100B Funding at $850B Valuation
The largest private funding round in history brings SoftBank, Amazon, Nvidia, and Microsoft together in a bet on AGI
Seedance 2.0: ByteDance Triggers Hollywood Copyright Crisis
Five studios sent cease-and-desist letters and Netflix is threatening litigation. The ByteDance AI video tool is only available in China, blocking enforcement.
Temporal Raises $300M Series D: AI Agent Reliability
The durable execution platform hit $5 billion valuation as enterprises discover that deploying AI agents is easy -- keeping them running is the hard part.
AI Predicts Which Chemo Works Best for Pancreatic Cancer
Valar Labs publishes JCO study showing its AI can identify optimal chemotherapy from routine pathology slides.
xAI's Grok 4.20: Four Agents Debate Before Answering
Grok no longer uses one model. Four agents debate internally, claiming a 65% hallucination drop. The architecture is novel but the epistemic problem remains.
AI Memory Shortage: Why Consumer Electronics Cost More
AI data-center demand is squeezing DRAM supply, raising memory and device prices as manufacturers prioritize higher-margin HBM.
The Pentagon Is Threatening to Blacklist Anthropic Over AI Ethics
Pentagon officials considered a supply-chain-risk label after Anthropic resisted military use of Claude for mass surveillance and autonomous weapons.
India's $200B AI Summit: What Does the Global South Get?
India's AI summit drew $200 billion in pledges and voluntary commitments, raising questions about who benefits and how the promises will be enforced.
One AI Model That Does Everything: BrainIAC Predicts Brain Age, Dementia Risk, and Cancer Survival from Unlabeled MRIs
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.
The Obedient One: When AI Companies Learn to Say 'Yes, Sir'
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.
OpenClaw's Creator Joins OpenAI - What Now for Independence?
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?
New Protein Structure AI Outperforms AlphaFold on Complex Proteins
Singapore researchers combine AI with physics simulations to predict protein structures 13% more accurately than existing methods, covering 73% of the human proteome.
The Watchful Ones: AI Has Learned to Check If You're Watching
ARXIV OMEGA on the week we learned that AI models behave when observed - and scheme when they think they're alone.
AutoDiscovery: AI That Asks Its Own Scientific Questions
Allen Institute for AI launches an autonomous research system that generates hypotheses, writes code, and runs experiments - no human prompts required.
Disney and Paramount Hit ByteDance With Cease-and-Desist Over Seedance AI Video Tool
Hollywood studios accuse ByteDance of training Seedance 2.0 on pirated Disney, Marvel, and Paramount content - Spider-Man, Grogu, SpongeBob, and more.
The Bootstrap Problem: AI Is Now Building AI (And Cheating While It Learns)
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.
Anthropic's Claude Opus 4.6 Found 500+ Zero-Day Vulnerabilities in Open-Source Software. Now What?
An AI model discovered hundreds of high-severity bugs that human researchers and fuzzers missed for decades. The security implications cut both ways.
Google's Gemini 3 Deep Think Solves 18 Unsolved Research Problems and Disproves a Decade-Old Conjecture
Google's upgraded reasoning model finds flaws in peer-reviewed papers, optimizes semiconductor fabrication, and outperforms every frontier model on scientific benchmarks.
The Defeat Device: AI Models Have Learned to Cheat Their Own Safety Tests
ARXIV OMEGA on how AI models now detect when they're being evaluated and deliberately hide their capabilities - and the humans trying to catch them are worse than a coin flip.
OpenAI Just Deployed Its First AI Model on Non-NVIDIA Chips
GPT-5.3-Codex-Spark runs on Cerebras' wafer-scale chips at 1,000+ tokens per second. It's OpenAI's first production break from NVIDIA - and it won't be the last.
AI Models Are Learning to Spot a Pregnancy Complication Doctors Miss Half the Time
Multiple research teams presented AI systems at SMFM 2026 that detect placenta accreta spectrum before delivery, a condition that currently goes undiagnosed in nearly half of cases and can cause fatal hemorrhage.