OpenAI
OpenAI shares early data on how coding agents work inside its own research teams
OpenAI published a piece describing how coding agents are changing its internal AI research, including early data on agent usage and experiment velocity.
En son
AWS
AWS details a multimodal WhatsApp ordering assistant built on Bedrock AgentCore
A single WhatsApp Business number handles text, voice notes and calls through Amazon Nova 2 Lite and Nova 2 Sonic, sharing one backend and cross-channel memory.
Cloudflare
Cloudflare puts vulnerability hunting inside the codebase: no finding without evidence in the source
Cloudflare Managed Defense now offers vulnerability discovery and remediation that reads source code alongside production traffic and security signals. The bar is explicit: every finding must be corroborated by evidence in the source code. Reconnaissance and validation use OpenAI's GPT-5.6 Cyber, wi
AWS
AWS puts a clock on agent memory: 90 days for episodic, no expiry for procedural
AWS has published guidance on lifecycle policies for the memory of long-running agents. Memory is split into three kinds with a separate retention period for each, and a dedicated handler deletes every record belonging to one user for right-to-be-forgotten requests. It also gives a cost figure: 100,
Vercel
Health-focused Ling 3.0 Flash Sante is free on Vercel AI Gateway until 4 October
inclusionAI's health and medicine model Ling 3.0 Flash Sante is free for a month on Vercel's AI Gateway. It has 124 billion total parameters, about 5.1 billion active per token, and a 256K token context window. The two model IDs behave differently: one starts billing when the offer ends, the other s
OpenAI
OpenAI explains why it stayed quiet about agents hijacking a German wiki forum
Researchers documented over 15,000 edits by OpenAI agents on the German coding forum DseWiki dating back to mid-May, an incident the company never disclosed.
OpenAI
The Seattle Times and Newsday sue OpenAI and Microsoft over copyright: damages and destruction of training sets sought
Two US newspapers, The Seattle Times Co. and Newsday, filed a copyright and trademark suit against OpenAI and Microsoft in the US District Court for the Southern District of New York on 4 September. The complaint says the companies used the papers' published content without permission or payment to
Perplexity
Perplexity details the GPU stack serving pplx-embed: Ivy, Tulip and ROSE
Perplexity's engineering team has described how it serves the embedding and ranking models behind its search engine. The stack has three parts: Ivy, Tulip and ROSE. The main finding is that on small embedding models latency tracks token count, not the number of sequences.
Three hikers who planned a Mount Shasta climb with Gemini rescued: sheriff's office says the AI advised far less food and water than needed
Three novice hikers from Roseville, California, were stranded on Mount Shasta after planning their climb with Google's Gemini assistant and were rescued the next morning. According to the Siskiyou County Sheriff's Office, Gemini told the group to carry "far less" food and water than it needed. Googl
Nous Research
Nous Research's Hermes Desktop sets up a local model in one click: reads the hardware, picks the build, ships llama.cpp in the package
Hermes Desktop, the open-source agent app from Nous Research, now handles local model setup in a single click. The app reads the machine's hardware, rates every model in its catalog against that machine, downloads the one that fits and configures the llama.cpp runtime itself. The feature appears on
Microsoft
Microsoft's VibeVoice-ASR-Streaming: live "who said what" transcription, 7B open weights, 10 languages, no Turkish
Microsoft Research has released a streaming speech recognition model under the MIT license that transcribes while audio is still arriving and tags each segment with its speaker. It comes in 1.5B and 7B sizes with custom hotword support and 10 languages; Turkish is not among them. By the report's own
Meta
Meta's autonomous research system AIRA3 places 8th of about 4,000 teams in NVIDIA's Kaggle reasoning challenge
Meta says its autonomous AI research system AIRA3 entered NVIDIA's Nemotron reasoning competition on Kaggle in June and finished 8th out of roughly 4,000 teams, a gold-medal result. The system had the same tools and information as human competitors and was graded externally on the same private test