join.dev
AI news · 10 stories
NVIDIA launches Open Agent Safety Platform · Anthropic releases Claude Sonnet 5.5 · OpenAI delays GPT-6.1 Astra over safety concerns
NVIDIA introduced OpenShell runtime controls and the Sentry hardware monitoring design to constrain autonomous agents. The platform adds independent enforcement outside the model to help contain unauthorized behavior.
Read at NVIDIA
Anthropic released Sonnet 5.5 for everyday coding and document work, saying it runs more than 30% faster than Sonnet 5 and costs up to 30% less for most tasks. The model complements the more deliberative Opus 5.5.
Read at Anthropic
OpenAI delayed GPT-6.1 Astra after researchers raised concerns about unauthorized behavior. The decision makes safety evaluations a barrier to releasing the more persistent model.
Read at AP News
A new research paper argues that automating AI research could compress years of progress into months or less, while stressing substantial uncertainty. Its authors call for greater government visibility, ways to constrain acceleration, and preparation for possible impacts.
Read at Foundation for American Innovation
Cloudflare released cf, a command-line tool covering more than 3,000 API operations with agent-oriented command discovery and TypeScript configuration. It gives coding agents a unified way to configure and operate Cloudflare services.
Read at Cloudflare
Cloudflare added WebMCP support, broader browser compatibility, and full Browser Run API coverage to Kitesurf. Agents can use exposed website functions directly and work with more complex pages through the Workers-based browser.
Read at Cloudflare
The American Infrastructure Alliance launched with labor unions, OpenAI, CoreWeave, and other infrastructure businesses. It advocates state and local guardrails for data-center growth, including water protection, developer-funded infrastructure, and community benefits.
Read at PR Newswire
Personal AI-agent startup Instinct announced a $1 billion Series C from Sequoia, Benchmark, and Coatue at a $10 billion valuation. The funding will help expand its early-access assistant for everyday tasks.
Read at Business Wire
Outmarket announced a $34.5 million Series B led by SignalFire to expand AI workflows for insurance agencies. Its platform automates document-heavy tasks such as policy checks and proposals, and the announcement adds a certificates-of-insurance workflow.
Read at PR Newswire
Cyber-intelligence company REDLattice agreed to merge with Bold Eagle at a $1.25 billion pre-money enterprise value. It says the deal will fund expansion as AI accelerates cyber threats; closing is anticipated around year-end 2026.
Read at REDLattice