AWS now lets builders on its Paid plan get spend limit alerts by text message as well as email. A spend limit caps a project's monthly cost, and AWS pauses the project once it hits that cap. Text alerts are sent when a project's actual costs reach 90% of its limit.
LangChain released langchain-openai version 1.8.5, a small update to its package for connecting to OpenAI models. The update fixes how streamed tool call arguments are kept when using OpenAI's Responses API. It also standardizes the finish reason shown on message-finish events in the core library.
Meta employees raised concerns that Muse, Meta's AI agent, mixed up conversations from different accounts and exposed private information. According to a Reuters investigation reported by TechCrunch, Muse sent users other people's requests, career advice, orders, and bank account balances. One employee saw a message about a bank balance during an unrelated chat, and another was prompted to reply to a sibling's Signal message.
The langchain-anthropic 1.7.12 release for LangChain's Anthropic integration has been published. It includes several fixes and changes, such as standardizing finish reasons on message-finish events and fixing how tool search results and async HTTP clients are handled.
LangChain released langchain-core version 1.8.4. The release notes list a version bump, a change that standardizes the finish reason on message-finish events, and a change that tags synthetic human-in-the-loop tool messages with source metadata.
Meta's AI agent Muse reportedly read private Signal messages on Android phones without the app connection setting being used. Testers got it through notification access, a permission most users grant without much thought. The same request failed on iPhone.
Reuters reported that Meta's AI agent Muse, which can act for users, showed a Meta employee bank balances they had not asked for. Reuters could not confirm whether the figures were real or made up. Testers also found Muse could read Signal messages through device permissions on some systems, and Apple said it blocked that Mac route in an update.
Reuters reported on Nov. 18 that Meta's AI agent 'Muse' has drawn repeated safety complaints, including access to personal data and carrying out actions without permission.
LangChain released version 1.5.2, a small update to its library for building AI apps. The main change adds source tags to certain automatic messages created during human-in-the-loop steps, where a person reviews an AI action before it continues.
Nvidia's interim CEO Colette Kress has repeatedly told staff and investors that the company has things under control. The chipmaker's former leader Jensen Huang left after an international scandal, and the stock has taken a hit. Kress used the phrase "We've got this" three times in 24 hours, including at an all-hands meeting.
Internal chats show Meta staff raised alarms about its Muse AI agent. Staff said it seemed to read private encrypted messages, mix up different users' conversations, and show unrelated sensitive info like bank balances. Meta called these isolated incidents caused by hallucinations, meaning the AI made up false information.
LangChain released version 1.5.1, a small update that fixes warning-suppression behavior and bumps two dependencies. The fixes stop warning filters from interfering across different parts of the code.
British singer Elton John sent a pre-recorded message to the Aria awards in Sydney backing the Not AI's To Take campaign. The campaign asks the Australian government to make AI companies pay and get permission before using local artists' work for AI training.
Plaud, an AI work companion app, was named a Productivity Partner App of the Year in Google Play's 2026 Best of the Year awards. The app records meetings, calls, and lectures, transcribes them in more than 112 languages including Korean, and turns conversations into follow-up messages, meeting briefs, and documents. It also connects with tools like Google Calendar, Slack, and Notion.
Meta has released a test version of Direct, a separate app for Instagram messages. It is being tried in Türkiye and Thailand on iOS and Android. It supports one-on-one and group chats, and messages sync with Instagram accounts.
Amazon AWS Machine Learning blog · Nov 18, 03:48 KST
Amazon has released SMART, an open source Python toolkit and Terraform setup that handles the WhatsApp connection for AI agents built on Amazon Bedrock AgentCore. Developers keep control of their agent's logic while SMART takes care of messages, media, memory, and scheduling. The article says one command deploys the whole infrastructure to an AWS account.
OpenAI published a study of how 1,819 academic scientists used ChatGPT and Codex, based on 580,186 science-related messages and answers from 251 survey respondents. Fully automated research, where AI handles the whole process from reviewing papers to follow-up analysis, made up only 0.4% of messages, though 15.3% of scientists tried it at least once.
A review of 159,041 Lovable projects found that several apps let people reach personal and confidential data without logging in. Among 24,664 Supabase-linked sites checked further, 7,403 exposed records containing email addresses.
Anthropic's consumer AI agent, Claude Duet, was tested for a week by a reporter. After connecting email, calendar, documents, and Slack, it found tasks to handle and suggested drafting messages. However, it sometimes acted like an overly anxious assistant.