News about foundation models, LLMs, multimodal models, benchmarks, and model releases from OpenAI, Anthropic, Google, Meta, Mistral, and others.
981 articles
A networking vulnerability in the OpenClaw project has been found to allow attackers to poison the underlying LLM through manipulated network traffic. Researchers warn the flaw could be exploited to corrupt model behavior or outputs. Source: darkreading.com
Security researchers have identified InjecMEM, a new type of attack targeting the persistent memory features of large language models. The technique could allow attackers to manipulate or corrupt stored context used by LLM-based systems. Source: startuphub.ai
A paper published in Nature Human Behaviour argues that despite appearing emotionally expressive, LLMs do not possess genuine feelings. The research examines why AI systems can convincingly simulate emotional responses without any underlying subjective experience. Source: nature.com
Traditional application security testing tools, such as SAST and DAST, fail to fully address vulnerabilities unique to LLM-based applications. The OWASP LLM Top 10 outlines these gaps and highlights the specific risks security teams need to watch for. Source: scworld.com
Thomson Reuters has launched Thomson, its first proprietary large language model designed specifically for legal work. The model combines the company's extensive legal content library with LLM technology. Source: siliconangle.com
Thomson Reuters has officially launched Thomson, a proprietary LLM trained on its Westlaw and Practical Law content. The move had been anticipated for months, with the company's CEO previously hinting at the project in earlier interviews. Source: lawnext.com
Travelers built TravelersLLM to handle insurance-specific queries in-house, reserving costly frontier models for complex reasoning, research, and coding tasks. The approach lets the company cut overall AI expenses while keeping routine queries efficient. Source: ciodive.com
New platforms launched by Stripe and Perplexity aim to make token consumption more cost-efficient. The initiatives reflect a broader industry push to optimize how AI usage costs are managed. Source: washingtonpost.com
Thomson Reuters announced on August 24 the launch of Thomson, the company's first in-house large language model built for the legal sector. Source: legaltechnology.com
Google researchers have long focused on equipping LLMs with harnesses so they can act more effectively as agents in complex environments. Now the same approach is being extended to the environments themselves, aiming to improve how agents interact with their surroundings. Source: eu.36kr.com
The European Union is evaluating whether ChatGPT should be classified as a very large online search engine under its digital rules. A ruling is expected soon and could bring OpenAI's chatbot under stricter EU oversight. Source: table.media
Israeli scientists found that large language models like ChatGPT and Gemini can analyze user behavior deeply enough to construct personality assessments. Their study suggests these AI systems can also anticipate how people are likely to respond in certain situations. Source: jpost.com
Alibaba completed a record share sale in Hong Kong, raising roughly $10 billion to accelerate its transformation from e-commerce giant into an AI powerhouse. The company plans to funnel the funds into building out its full-stack AI infrastructure and capabilities. Source: freemalaysiatoday.com
Lilly Ryan, a cybersecurity engineer at Thoughtworks, argues that AI's real impact on cybercrime is speed, not novel 'magic' malware. She explains how AI lets attackers scale existing techniques far faster than human-driven methods, and what companies should do in response. Source: techobserver.in
Researchers have developed LLM4M, a domain-specific large language model designed to support maintenance decision-making for wind farms. According to the study, the model helps streamline planning processes tailored to the unique operational needs of wind energy infrastructure. Source: eurekalert.org
The new V4 Flash Vision Exp model is initially available only through the Chinese startup's paid developer platform. DeepSeek has hinted a free version could follow later. Source: siliconangle.com
Anthropic's newest large language model, Claude Opus 5, is facing significant backlash as users report various performance issues since launch. Complaints span multiple aspects of the model's behavior and reliability. Source: startuphub.ai
Mistral, one of Europe's leading AI companies, is offering access to Z.ai's GLM-5.2 model, exposing a tension between European sovereignty ambitions and commercial reality. The move shows how business incentives can clash with political goals in the AI race. Source: scmp.com
NEC has begun building what it calls an LLM for the ocean, training a foundation AI model on massive amounts of underwater data. The goal is to create an AI system specialized in understanding marine environments. Source: splash247.com
According to Volcano Engine, Tesla began deploying the Doubao large language model in its vehicles on August 19. The update is being rolled out gradually to Tesla's infotainment systems. Source: moomoo.com