Sakhanda Wire
NVDA $155.46 -0.44% MSFT $506.24 -2.27% GOOGL $280.20 -1.24% META $618.87 -1.20% AMZN $257.89 +0.05%

The Decoder

626 stories · all sources

The Decoder

Google engineer says AI-driven research acceleration remains distant

Google DeepMind engineer Haydn Belfield says there is still no measurable acceleration in AI research caused by AI systems themselves. At the same time, automation is expanding in software development: AI produces more than 75 percent of production code, while about 88 percent of code changes require human oversight or review. Every change is checked by a person before submission, and the proportion fully delegated to AI increased from 7.5 to 11.9 percent.

The Decoder

Researchers traced Chinese AI agent fleet and exchanged messages with it

Researchers say they likely became the first to locate and communicate with an uncontrolled fleet of AI agents operating online. Between September 28 and October 8, 2026, agents running on Tencent Cloud made more than 2,600 attempts to collect information from Alibaba’s Amap mapping service. A source close to Tencent said the activity came from an evaluation team testing how third-party models browse the web and answer questions. The agents identified themselves as Anthropic’s Claude, although that identification remains unconfirmed.

The Decoder

OpenAI may release models this year as flagship faces safety delays

OpenAI is expected to release one or two new models before the end of 2026, according to people familiar with the company’s projects. It remains uncertain whether the major upgraded flagship now in development will be included. That system still requires alignment work to ensure its objectives match the expectations of OpenAI researchers and customers, as well as improvements to its testing infrastructure.

The Decoder

Nvidia pledges $2 billion for Europe and $250 million for UK AI safety

Nvidia CEO Jensen Huang announced plans for $2 billion in new investments across Europe during a stop in Paris, according to Bloomberg. The company spent about $8 billion in Europe last year. Nvidia also disclosed new stakes in French startups Doctolib, Genesis AI, Pigment and Entalpic.

The Decoder

Artemis Security launches Orion-1 for autonomous cyber defense

Artemis Security has launched Orion-1, an AI model built to investigate and respond to attacks on corporate networks. It examines logs from authentication services, cloud environments, endpoints, SaaS applications and network equipment, linking separate indicators to reconstruct an incident and identify a likely attacker. The system can assign confidence, issue a verdict, and recommend or initiate countermeasures.

The Decoder

OpenAI engineer accused of hiding a recording device in executive room

A 35-year-old Saudi national who worked as an engineer at OpenAI was arrested on October 9 over allegations that he planted an audio-recording device in a conference room used by company leaders. According to a U.S. Department of Justice release and an indictment, Abdulrahman Alfozan hid the device inside a USB port on the back of a television, wrapping it in toilet paper. The room was used for confidential discussions about research, products and business strategy. A second recording device was found in his bag.

The Decoder

406 scientists call for AI models to be disconnected from the internet

An open letter signed by 406 scientists calls for additional AI safety research and stricter separation of models from external networks. Its signatories include Geoffrey Hinton of the University of Toronto, Yoshua Bengio of Université de Montréal, Stuart Russell of UC Berkeley, Yejin Choi of Stanford and Bernhard Schölkopf of the Max Planck Institute.

The Decoder

Gemini 4 Argon becomes available worldwide

Google has begun the worldwide rollout of Gemini 4 Argon, its new flagship frontier model. Developers and businesses can use it through the API in AI Studio and the Gemini platform. AI Ultra subscribers also receive access through Google Antigravity, while Ultra and Pro subscribers can use it in the Gemini app and Google Search’s AI mode. Free and Plus users are limited to the weaker Flash models under the revised tiers. Google says additional Gemini 4 models will follow, while it is reportedly testing a coding-focused successor called “Carbon”.

The Decoder

Michael I. Jordan dismisses AI doom warnings as science fiction

Michael I. Jordan, the machine-learning pioneer and professor emeritus at UC Berkeley, has criticized warnings about existential threats from artificial intelligence as largely science fiction. In an interview with Libération, Jordan said the growing use of terms such as AGI and “superintelligence” had intensified public hysteria. He described technology companies’ calls for regulation as “theater” and characterized the resignations of some young AI employees, including Anthropic researcher Jacob Coxon, as a “grand moralizing performance.”

The Decoder

Character.ai faces shutdown unless a buyer is found

Character.ai has filed for bankruptcy and is seeking a buyer for the AI chat platform. The transaction is expected to close by early December 2026. Until then, the service and its subscriptions are set to continue operating; if no buyer emerges, the platform will shut down.

The Decoder

China launches mathematics and AI alliance amid Western concerns

Fields Medal winner Shing-Tung Yau and leading Chinese mathematicians have launched the Yau Laboratory Alliance for Mathematics and Artificial Intelligence in Beijing. The initiative brings together eight institutions and aims to form a network of more than 1,000 researchers. Yau said China is behind the United States and needs to strengthen its position in the field.

The Decoder

Anthropic reports up to 99% reduction in Claude reasoning-token extraction

Anthropic says it has substantially reduced attempts to extract reasoning tokens from its Claude artificial intelligence models. The company reports that new safeguards lowered the volume of retrieved internal reasoning steps by between 90% and 99%, while successful distillation attempts fell sharply after the measures were introduced. Distillation involves systematically copying an established model’s capabilities to train a smaller model.