The live wire

Latest News

Chronological reporting and platform updates, with reporting labels kept explicit.

Editorial feed

Published records

ModelsJul 16, 2026

Moonshot previews Kimi K3, with full weights due July 27

Moonshot AI has published documentation for Kimi K3, describing a 2.8T-parameter multimodal model with a 1M-token context window and saying full weights are planned for July 27.

3 min readChecked Jul 18, 2026Source checked

Source: Moonshot AI — Kimi K3 quickstart

Why this matters: K3 raises the ceiling claimed by open-weight model developers, but independent testing has to wait until the promised weights arrive. The gap between a preview and a reproducible release matters.

Read story →

PolicyJul 16, 2026

OpenAI outlines its case for teen access paired with stronger safeguards

OpenAI has published its case for allowing teenagers to use AI while applying protections designed for their age and stage of development.

3 min readChecked Jul 18, 2026Source checked

Source: OpenAI — Why teens deserve access to safe AI

Why this matters: Teen access is becoming a product-design and policy question, not only a parental choice. The effectiveness of the described safeguards still needs independent evidence and sustained scrutiny.

Read story →

RoboticsJul 16, 2026

Fujitsu and Japanese robotics leaders explore a shared physical-AI platform

Fujitsu says it is exploring a collaborative-control platform with FANUC, Yaskawa Electric, and Kawasaki Heavy Industries, using NVIDIA open physical-AI technologies.

3 min readChecked Jul 18, 2026Source checked

Source: Fujitsu — official press release

Why this matters: Industrial physical AI will depend on robots from different manufacturers coordinating safely. A shared control layer could be consequential, but this is an exploration of business opportunities—not a finished product launch.

Read story →

ModelsJul 15, 2026

Thinking Machines launches Inkling as a customizable open-weights model

Thinking Machines has released Inkling, an Apache 2.0 open-weights multimodal foundation model positioned as an efficient, customizable base rather than the strongest model overall.

3 min readChecked Jul 18, 2026Source checked

Source: Thinking Machines — Introducing Inkling

Why this matters: Inkling tests a different product thesis: a model’s value can come from how effectively teams can adapt it, not only from its position on general-purpose benchmarks.

Read story →

PolicyJul 15, 2026

Australia sets out a national AI framework and new standards

Australia’s government has announced that it will introduce an Australian AI framework and standards, building on its existing Data Centre Expectations.

3 min readChecked Jul 18, 2026Source checked

Source: Prime Minister of Australia — AI in Australia’s interests

Why this matters: National AI rules increasingly combine industrial strategy with safety expectations. The details will determine whether the framework changes deployment practice or mainly coordinates existing policy.

Read story →

ResearchJul 15, 2026

OpenAI details GPT-Red, an internal model built to attack other AI systems

OpenAI has described GPT-Red, an internal red-teaming model trained through self-play to generate adversarial attacks against other AI systems.

3 min readChecked Jul 18, 2026Source checked

Source: OpenAI — Unlocking self-improvement with GPT-Red

Why this matters: Automated attackers can widen security testing beyond hand-authored prompts, but the results are OpenAI’s own evaluation. Reproducibility and transfer to other systems remain important questions.

Read story →

ApplicationsJul 14, 2026

Anthropic introduces free Claude access and teaching tools for verified US educators

Anthropic has introduced Claude for Teachers, offering verified K–12 educators in the United States free access to premium Claude capabilities and curriculum-linked teaching resources.

3 min readChecked Jul 18, 2026Source checked

Source: Anthropic — Introducing Claude for Teachers

Why this matters: Free access could accelerate classroom adoption, but schools still need local policies for privacy, accuracy, student use, accessibility, and human review of instructional material.

Read story →

ModelsJul 9, 2026

OpenAI releases the GPT-5.6 model family across ChatGPT, Codex, and its API

OpenAI has released GPT-5.6 across ChatGPT, Codex, and the OpenAI API, publishing product availability, pricing, safety information, and company-run evaluation results.

4 min readChecked Jul 18, 2026Source checked

Source: OpenAI — GPT-5.6

Why this matters: The release changes the default capability frontier for many OpenAI users. Its benchmark and efficiency claims are vendor-reported and should be separated from independent comparative testing.

Read story →

ModelsJul 9, 2026

Meta launches Muse Spark 1.1 and previews a developer model API

Meta has introduced Muse Spark 1.1, made it available in Meta AI, and opened a public preview of the Meta Model API for developers.

3 min readChecked Jul 18, 2026Source checked

Source: Meta AI — Introducing Muse Spark 1.1

Why this matters: A new developer API gives Meta a more direct role in the agent-model market. Performance and safety claims in the launch material remain Meta’s own and need outside evaluation.

Read story →

ResearchJul 9, 2026

Google Research presents SensorFM for wearable health data

Google Research has presented SensorFM, a foundation model for wearable sensor data that researchers report transfers across 35 prediction tasks.

3 min readChecked Jul 18, 2026Source checked

Source: Google Research — SensorFM

Why this matters: A general representation for wearable signals could reduce task-specific model building. Clinical usefulness, subgroup performance, privacy, and independent validation still need careful scrutiny.

Read story →

Published records

The important AI developments. None of the noise.