Aim Labs discloses zero-click Microsoft 365 Copilot vulnerability EchoLeak enabling data exfiltration
Aim Labs discovered and disclosed to Microsoft MSRC a zero-click vulnerability chain in M365 Copilot called EchoLeak,…
AI news monthly archive: May 2025 features 34 curated AI news items covering model releases, product updates, industry news, papers, tutorials and opinions, each with an AI summary, source credit and original link.
Aim Labs discovered and disclosed to Microsoft MSRC a zero-click vulnerability chain in M365 Copilot called EchoLeak,…
Modal announced that B200 and H200 GPUs are available to all accounts without contacting sales. B200 is priced at…
🚀 DeepSeek-R1-0528 is now released! 🔹 Improved benchmark performance 🔹 Enhanced frontend capabilities 🔹 Reduced…
DeepSeek released DeepSeek-R1-0528, a minor version upgrade of the R1 model, which achieves leading results among all…
OpenRouter has launched Reasoning Streams, which supports streaming summaries of reasoning processes, along with…
DeepSeek upgraded the deepseek-reasoner model to DeepSeek-R1-0528, significantly enhancing reasoning capabilities,…
Mistral AI has released the Agents API, a dedicated framework for agent use cases that complements the Chat Completion…
Hugging Face released Tiny Agents in Python, inspired by Tiny Agents in JS, extending the huggingface_hub client SDK…
Successful enterprise AI deployment rests on three pillars: leadership ensures strategic alignment, labs incubate…
The MCP Agent Server example demonstrates an "Agent-as-a-Service" architecture that wraps agent workflows as MCP…
Cognition released the official DeepWiki MCP Server, free and requiring no login or authentication, enabling…
TII's Falcon LLM team released the Falcon-H1 series, comprising six open-source models (0.5B, 1.5B, 1.5B-Deep, 3B, 7B…
The Hugging Face team released nanoVLM, a minimal toolkit repository for training vision-language models in pure…
Hugging Face has published a guide to Diffusers quantization backends, comparing how bitsandbytes, torchao, Quanto,…
Mistral AI and All Hands AI have jointly released Devstral, an agentic coding model for software engineering tasks,…
DreamGen lets robots "dream" synthetic training data in video generation models. By fine-tuning models such as Sora to…
The Gemma 3n preview is released, an open-source multimodal model optimized for mobile devices. It uses a 2-in-1…
Google plans to expand Gemini into a world model that can plan and imagine new experiences by simulating the world,…
Google released its next-generation generative media models Veo 3 and Imagen 4, along with Flow, a tool built for…
ARC Prize released the ARC-AGI-2 technical report, introducing a new benchmark for frontier AI reasoning systems with…
In an appendix to the o3 and o4-mini system card, OpenAI introduced Codex, a cloud-based coding agent powered by…
TII's Falcon-Edge series is based on the BitNet architecture, pre-trained directly with ternary weights {-1, 0, 1},…
Hugging Face officially announced that Transformers will serve as a cross-framework standard for model definitions,…
Cognition released Devin 2.1, adding 🟢🟡🔴 confidence ratings, where confidence scores correlate strongly with task…
LangChain announced the general availability of LangGraph Platform, an infrastructure and management layer for…
LlamaIndex has released an improved Memory component that provides agents with short- and long-term memory. In its…
Prime Intellect released INTELLECT-2, which it calls the first 32B-parameter reasoning model trained through global…
Proposes the "Physical Turing Test" as the north-star goal for general robots: whether a machine can complete physical…
Mistral AI has released Mistral Medium 3, positioned to deliver near-flagship performance at about 1/8 the cost,…
Citing positive developer feedback, Google released an updated Gemini 2.5 Pro Preview two weeks early, with further…
Cognition launched DeepWiki, making Devin Wiki and Devin Search freely available. By replacing github.com in a URL…
OpenAI has published a blog post detailing what it missed on the model sycophancy issue, describing the problems…
Sycophancy in large language models reveals the nature of their "personality" and persuasion mechanisms. By observing…
Suno released its latest music generation model, v4.5, calling it its most expressive version yet. Style options are…
Sponsored
Orange Relay (pro.nat6.net) is a premium AI API relay — an excellent transfer station with competitive pricing, reliable quality and broad model coverage including Claude, GPT and Gemini.
Visit Orange Relay