Morning Edition · Tuesday, July 14, 2026
Tech
Meta Opens a Paid Model API With Muse Spark 1.1, Its Turn Toward Metered Access
The Superintelligence Labs model ships a one-million-token context window and agentic tool use, priced at $1.25 per million input tokens and $4.25 per million output tokens.
Tech
Anthropic Restores Fable 5 After a Cyber Jailbreak Forced It Offline
The model returned globally on July 1 after a classifier blocked more than 99% of the identified attack, and Anthropic proposed a cross-lab framework for scoring jailbreak severity.
Tech
Meta's Non-Invasive Brain-to-Text Decoder Reaches 78% Word Accuracy for Its Best Subject
Brain2Qwerty v2 reaches a 39% average word error rate from magnetoencephalography alone and removes the earlier requirement to know keystroke timing in advance.
Tech
Bilibili Releases Index-1.9B, a Small Open Model Trained on 2.8 Trillion Tokens
The four-model series carries 1.9 billion non-embedding parameters and targets a size that runs on consumer hardware, extending China's downloadable-weight stack.
Tech
Researchers Bypass MedGemma-4B's Safety Rules With Trivial Prompt Reframing
A new paper shows Google's open-weight medical model issues exact drug dosages and definitive diagnoses its model card forbids, using only rephrased prompts.
Tech
Meta Ships Muse Image and Muse Video With Editable, Agentic Generation
The Superintelligence Labs image model plans, searches the web, and critiques its own drafts, and it accepts sketch-on-image edits rather than only text prompts.
Tech
Paper Shows Prompt Formatting Alone Can Flip Language-Model Leaderboards
A token-controlled study introduces a Format Sensitivity Index measuring how much identical questions in different wrappers move model scores.
Markets
Japan's JX Metals Reports 90% Lithium Recovery From Spent EV Batteries
A plant in Tsuruga nearly doubles the prior sub-50% industry rate by swapping recovered lithium hydroxide for sodium hydroxide, with mass production set for 2027.
Tech
Study Finds Message Format Effects in Multi-Agent Handoffs Depend on Model Tier
A paper reconciles two conflicting literatures, reporting that whether structured messages help or hurt agent relays varies with model capability.
Tech
Position Paper Argues Every Ground Truth Is a Human Construction
The authors challenge the treatment of labeled datasets as neutral objective measurements, with consequences for how models are trained and evaluated.
Tech
AuditWeave Proposes a Tamper-Evident Evidence Layer for AI-Assisted Decisions
The system targets regulated domains where an organization must later reconstruct which AI steps produced a consequential decision.
Tech
CLIR-Bench Tests Multimodal Models on Irregular Clinical Time Series
The benchmark targets the sparse, asynchronous, unevenly sampled measurements that dominate real patient monitoring and defeat standard sequence models.