AM-IA brings together news, research papers and applications relating to artificial intelligence published around the world into a single feed — and makes them available in your European language at the click of a button.
Scroll to browse · Click to filter
Forty times larger than the first Mercury but just as fast, Inception’s new diffusion model claims to outperform GPT-6 Luna and Claude Haiku 5.5.
Signed in June but made public on 19 October, an order from the Ministry of Internal Security requires DeepSeek to be removed, including where it is integrated into third-party products.
Notion’s sixth-generation agent infrastructure: task completion times have increased from 5 minutes to over 24 hours, and a router promises results on a par with the best models at 90 per cent less cost.
Out of around 10,000 decisions and 15 US cases, false information misleads LLM agents far more effectively than traditional prompt injection: a 62 per cent success rate, compared with 14 per cent.
Without ever touching the robot, an attacker can crush or release an object by sticking on a thin magnet label. Success rate: over 95%.
By quoting EN 18286 in the Official Journal, Brussels offers high-risk AI suppliers a first recognised means of proving compliance with Article 17 of the Regulation.
Selecting mostly rainy episodes to train and test models can reverse the diagnosis: the same model seems too cautious or too alarmist depending on the test set.
An object detector adapted to damaged glass plates star and source markers to rescale their position on the sky, with a recovery rate of 97.3% and a two and a half times faster extraction.
Six animal species, more than 60 organs and 14,000 studies: a family of numerical pathology models dedicated to preclinical drug safety trials.
▶
Al Chen (Cisco) raconte comment faire passer un logiciel d'IA par les contrôles de sécurité et de conformité de cinq grandes banques, sans y perdre un an.
Microsoft is expanding Project Perception with ‘Guardian Agents’ and a Defender capable of identifying more than 60 types of agents, restricting their access rights and requiring human validation.
Major Canadian publishers accuse Anthropic of scraping their articles to train Claude, bypassing the technical safeguards on their websites. They had previously taken legal action against OpenAI in 2024.
With TorchNeuron, AWS aims to neutralise Nvidia’s software advantage: a PyTorch application can switch to Trainium without any rewriting, and Amazon is contributing upstream to make PyTorch hardware-agnostic.
Symbolic reasoning, deep learning, similar case search and LLMs: by combining these approaches, FusionRare outperforms GPT 5.5 and DeepRare on 2,528 individual clinical cases, saving doctors 21 per cent of their time.
Twelve Xe-100 modular reactors with a combined capacity of 960 MW, operated by AEP near Terre Haute: Amazon has tripled the scale of its first agreement with X-energy, even though the company’s pioneering reactor in Texas has not yet been launched.
The Super VL, with 120 billion parameters, is aimed at enterprise users, whilst the Bolt 74B-A9B and E4B are designed to run locally on RTX Spark and DGX Spark machines.
Spun out of the University of Michigan, the start-up uses AI agents and Nvidia’s Cosmos simulator to automatically generate edge cases. It is already profitable thanks to autonomous driving.
Union Square Ventures, Khosla and Homebrew are leading Tolan’s Series B funding round; the app claims to have 250,000 active users, 70 million conversations in a year and recurring revenue in the tens of millions.
Researchers have published 1,000 X-rays of children in cardiac intensive care, annotated on a device-by-device basis, along with a tool that is already integrated in real time at Boston Children’s Hospital.
▶
Evan Kormos (Coinbase) explique comment faire entrer des agents dans des processus réglementés : modéliser les flux, fiabiliser les contrôles, puis passer de la supervision totale à la gestion des seules exceptions.