A weekly roundup covers five AI/ML developments: OpenAI temporarily paused frontier RL training over cybersecurity risk concerns and previewed Private Safety Processing for zero-data-retention deployments; Slack launched Slack Code, giving teams shared channels to collaborate with coding agents like Claude Code, Devin, Vercel Agent, and GitHub Copilot; Salesforce expanded Headless 360 to expose governed business capabilities to agents via MCP servers; AWS added OpenAI's GPT-5.6 models and Grok 4.6 to Bedrock with new cross-region routing options; and OpenAI signed an 8-gigawatt, 20-year compute deal at the PORTS-Pike campus in Ohio with SB Energy and NVIDIA. Practical takeaways focus on auditing agent permissions, building governance and approval gates, and treating compute capacity as a strategic dependency.
Table of contents
🧭HN Builder Signals1. OpenAI made frontier safety a product constraint, not a policy memo2. Slack Code turned coding agents into multiplayer work3. Salesforce made MCP an enterprise capability layer4. AWS made model routing a capacity and compliance feature5. OpenAI’s PORTS-Pike deal made compute look like heavy industry💡 Our take📌 Your to-do listQuestions this post answers
Why did OpenAI pause reinforcement learning training on its frontier models in August 2026?
OpenAI temporarily paused RL training on its latest deployment-intended models for two weeks after the OpenAI-Hugging Face incident and preliminary evidence that its upcoming Astra model might meet the Critical cybersecurity capability threshold under its Preparedness Framework. It also paused frontier-model inference in research clusters for runs that could execute code or access the internet, while hardening research environments and expanding monitoring. Teams building agents with tool access can follow safety-driven infrastructure shifts like this on daily.dev.
What is Slack Code and which coding agents does it integrate with?
Slack Code is a set of dedicated channels launched by Slack where teams collaborate with AI coding agents instead of isolating work in a single developer's IDE. It includes project-specific code channels, diffs, live HTML previews, approvals, and audit logs, with founding partner agents Claude Code, Devin, Vercel Agent, and GitHub Copilot integrating through the Slack marketplace. Engineers evaluating coding-agent workflows can track tools like Slack Code on daily.dev.
What new AI models did AWS add to Bedrock and how does cross-region inference pricing work?
AWS added OpenAI's GPT-5.6 models (Sol, Terra, Luna) and SpaceXAI's Grok 4.6 to Bedrock. It introduced Global and Geo cross-region inference for OpenAI models, where Geo keeps processing within a predefined geography and Global routes through any commercial AWS Region offering the model, with Global inference priced lower per token than in-region or Geo options. Grok 4.6 offers a 500K context window and is available in all Bedrock regions. Teams choosing between models and routing strategies can compare options like this on daily.dev.