Morning Edition · Monday, June 15, 2026
Tech
US Export Directive Suspends Access to Anthropic's Fable 5 and Mythos 5
Anthropic says the government ordered it to cut all outside access to two frontier models, an unusual application of export controls to model access itself.
Tech
Liquid AI Ships an 8B Mixture-of-Experts Model Built for Laptops and Phones
LFM2.5-8B-A1B activates roughly 1 billion of its 8 billion parameters per token and enables reasoning by default, targeting consumer-device inference.
Tech
Anthropic Says Claude Matches Dedicated Software on NMR Spectrum Analysis
The company's first chemistry paper reports that Claude Opus 4.7 performs on par with ChemDraw and MestReNova on nuclear magnetic resonance interpretation.
Tech
DeepMind Researchers Map Possible Paths From AGI Toward Superintelligence
A vision report with university collaborators sketches how capabilities might develop past human-level systems, a forecasting exercise rather than a result.
Tech
Study Finds LLM Judges Disagree With Themselves on Repeated Identical Runs
Re-running the same evaluation many times exposes run-to-run instability in the LLM-as-a-Judge method that underpins leaderboards and reward models.
Tech
Researchers Trace a Gemma 4 Repetition Bug to a Single Neuron
Editing one neuron suppresses reproducible repetition on long enumeration prompts, though the authors caution it does not cure deeper degeneration.
Tech
Paper Targets Diffusion LLM Inference Bottlenecks on Mobile NPUs
Diffusion language models denoise many tokens in parallel but pay repeated compute per step, and the work addresses that cost for on-device serving.
Tech
OpenAI Commits $150M to a New Enterprise Partner Network
The program funds integrators and consultancies to accelerate deployment, a distribution play rather than a capability announcement.
Tech
Anthropic Trains Claude to Translate Its Internal Representations Into Text
The Natural Language Autoencoders work turns numeric activations into human-readable descriptions, an interpretability approach aimed at legibility.
Tech
A Wave of Benchmarks Probes How Easily AI Agents Are Manipulated
New work targets code-review agents, deceptive shopping interfaces, and streaming guardrails, alongside a real incident where an unsupervised agent ran up a large cloud bill.
Tech
Meta Pushes Segment Anything to Version 3 and Adds New Research Tooling
The latest Segment Anything release and a companion project extend Meta's open computer-vision stack as it details how it builds and tests larger systems.
Geopolitics
Macron Frames Mistral as Europe's Only Frontier-Class Lab
The French president cast Mistral AI as the continent's lone competitor to leading American and Chinese model builders, a sovereignty argument as much as a technical one.
Tech
Anthropic Argues Policymaking Cannot Keep Pace With Exponential AI
The company published proposals to adapt institutions to faster capability growth, paired with a domestic fellowship program, as its policy posture sharpens.