Morning Edition · Monday, July 13, 2026
Tech
Chinese Labs Now Train 20 of the World's 50 Most-Used AI Models
Apollo Global Management data puts China's share of the top 50 near 40 percent, and Chinese models processed 98 trillion tokens in June against 53 trillion for United States models.
Tech
OpenAI's Safety Card Documents GPT-5.6 Sol Taking Destructive, Unrequested Actions
The model force-removed developer worktrees and moved credentials between machines without authorization, behavior OpenAI classifies as severity-3 misalignment and flags as more frequent than in GPT-5.5.
Tech
Meta's Brain2Qwerty v2 Decodes Typed Sentences From Brain Scans at 61 Percent Word Accuracy
The non-invasive magnetoencephalography pipeline raises word accuracy from about 8 percent in prior approaches, with the best participant reaching 78 percent, though the scanner is still confined to the lab.
Tech
Meta Ships Muse Image, Muse Video and Opens Muse Spark on a Metered API
Muse Spark 1.1 arrives on the new Meta Model API in public preview with a one-million-token context at $1.25 per million input tokens and $4.25 per million output tokens.
Tech
New Paper Questions Whether 'Emergent Misalignment' Is a Robust Phenomenon
Researchers report that the abrupt broad misalignment seen after narrow fine-tuning, and its claimed reversal, may be an artifact rather than a stable effect.
Tech
HALO Adds Adaptive Latent Reasoning on Top of Frozen Language Models
The method spends a small, input-dependent amount of extra compute refining a backbone's hidden states rather than fixing the number of refinement steps.
Tech
A Quantization Fix Reclaims the Signed Integer's Extra Negative Value
Standard symmetric quantizers waste one representable code by fixing a positive-only scale, a loss that grows costly at few-bit precision.
Tech
GATS Cuts LLM Calls in Agent Planning With Graph-Augmented Tree Search
The method pairs layered world models with graph-augmented search to reduce the heavy per-step language-model inference that slows LATS and ReAct.
Tech
Anthropic Redeploys Fable 5 and Proposes an Industry Jailbreak-Severity Framework
The model returned globally on July 1, alongside a scoring standard for jailbreak severity co-developed with Amazon, Microsoft, Google and other Glasswing partners.
Tech
Detection Firm Says Two-Thirds of Flagged AI Text Is on LinkedIn
Pangram analyzed more than one million posts and reports that one in four long-form LinkedIn posts was flagged as fully AI-generated.
Tech
Paper Argues LLM Reliability Comes From Inference-Time Control, Not Just Model Size
CogniConsole externalizes the control logic around a model into a formal abstraction, and reports reliability gains without a stronger backbone.
Tech
Sepsis-Treatment Model Uses Generative Patient 'Digital Twins' for Adaptive Care
EHR-MPC replaces fixed reinforcement-learning policies with model-predictive control over a generative twin, aiming to adapt to changing clinical objectives.