Stay informed. Keep your time.
8 stories · ~3 min readEstimated reading time for the entire edition.
Google has launched Gemini 4 Argon globally, offering it through the API, Gemini app, Search AI Mode, and Antigravity on selected subscription and enterprise plans. Google says the model ranks first on the Vals Index; the API has a combined input-output limit of 1 million tokens.
Gemini 4 Argon is available through the API and several Google apps to selected consumer and enterprise plans.
The study embeds encrypted neural networks in model weights to create backdoors that activate only with a hidden trigger. The authors report resistance to various detection methods on a small chemical-reaction prediction task; the security guarantees rely on cryptographic assumptions.
Researchers embedded encrypted neural networks in model weights to create capabilities activated by a hidden trigger.
Character.ai says it has begun a voluntary Chapter 11 bankruptcy process to facilitate a court-supervised sale of the company. It expects the sale process to conclude in late November or early December 2026; the platform will shut down if no suitable offer emerges. Services and subscriptions continue for now.
The company says the platform could shut down if no suitable sale offer emerges.
Anthropic announced measures against distillation attempts to extract Claude’s hidden reasoning traces, including restricting context transfer, using classifiers, and blocking repeat accounts. The company says extracted reasoning tokens fell by 10–100 times; it says the durability of these gains is uncertain and false positives remain below 0.1%.
Anthropic reports that methods making Claude’s hidden reasoning traces harder to extract cut the data collected by distillers by about 90–99%.
OpenAI engineer Abdulrahman Alfozan was detained on suspicion of placing a recording device in an executive meeting room at the company’s San Francisco headquarters and was charged with attempting to record communications. OpenAI said it used its models to examine the device and found no evidence that model weights, source code, or customer data had been taken.
An OpenAI engineer was detained on suspicion of secretly placing a recording device in an executive meeting room. The company said it used AI models to examine the device and found no evidence that sensitive data was taken.
Liquid AI released the open-weight LFM2.5-24B-A2B model for agents and tool use. The company reports a 131,000-token context window and a speed of 184 tokens per second on an M5 Max. In its own tests, the model is competitive on tool tasks but trails larger models in knowledge and coding.
LFM2.5-24B-A2B is intended for on-device agents and has a 131,000-token context window.
Google introduced Gemini 4 Argon as a general-purpose model for physical reasoning, building simulations, and controlling robots across the robotics lifecycle. The company says the model scored 87% on the ERQA benchmark using pure reasoning, outperforming other frontier models.
Gemini 4 Argon is presented as a general-purpose model for robotics simulation and manipulation tasks.
The article introduces Minions, a distributed virtual GPU cloud that combines idle Apple Silicon GPUs across macOS devices to run machine-learning inference workloads. The authors say it offers flexible task execution and fault tolerance; the summary provides no specific performance measurements.
Minions combines idle Apple Silicon GPUs to create a distributed virtual GPU cloud for machine-learning inference.
That’s it for today.The rest of your time is yours.
↑Stories are selected from Hacker News; accessible original articles are summarized and translated with AI. Summaries may contain errors; check the source beneath each story for details. Preparation starts daily at 06:55 Türkiye time (UTC+3), targeting publication before 07:00.
This edition covers articles shared on Hacker News on the date shown above.
Read or listen to a new edition every morning with Kısa for iPhone, save stories for later and keep your streak.
