i wake up, there is a louder week of AI updates.

Codex on the phone can now drive a Mac while the screen is off and locked.

Cursor is now available in Jira, with @mentions kicking off cloud agents that return merge-ready PRs.

Codex threads now follow the user between the ChatGPT mobile app and the desktop.

A new Keep this Mac awake toggle lets Codex keep running at the desk while the user works from elsewhere.

Cursor in Jira shipped to the changelog alongside the launch tweet.

Goal mode graduated from experiment and is now in the Codex app, IDE extension, and CLI.

Goal mode is positioned for tasks Codex can chase for hours or even days.

Cursor Automations moved into the Agents Window and got a week of 50 percent off agent runs.

Automations can now attach multiple repos so agents reason across codebases.

Automations can also run with no repos attached, opening the door to Slack digest style agents.

Appshots let users press Command-Command to attach an app window straight into a Codex thread.

Composer 2.5 launched as Cursor's most capable model, with included usage doubled for a week.

Composer 2.5 is pitched as up to ten times more efficient than similarly capable models.

Training scaled with more complex RL environments and text feedback used for credit assignment.

Composer 2.5 shares its open-source base, Moonshot's Kimi K2.5, with Composer 2.

Cursor and SpaceXAI are training a much larger model from scratch on roughly ten times the compute.

The Cursor SDK opened up in Python and TypeScript, with Composer usage 90 percent off for the long weekend.

Composer 2.5 landed in the changelog the same day as the announcement.

OpenAI and Dell partnered to bring Codex into hybrid and on-premises enterprise environments.

Gartner named OpenAI a Leader in its 2026 enterprise coding agents evaluation.

OpenAI Guaranteed Capacity offered customers long-term, reserved access to OpenAI compute.

Business workspaces can now distribute custom Codex plugins and manage internal tool availability.

Codex Analytics added user leaderboards, plugin usage, and finer token and run breakdowns.

Claude introduced Compliance API integrations so IT teams can govern it like other stack apps.

Project Glasswing has found more than ten thousand high or critical severity vulnerabilities since launch.

Anthropic argued the industry will need to adapt to the volume of vulnerabilities models like Mythos Preview can find.

OpenAI published a content provenance update covering its broader transparency push.

ChatGPT images now carry C2PA credentials plus a SynthID watermark, with a public verification tool.

Claude Managed Agents launched self-hosted sandboxes in public beta and MCP tunnels in research preview.

MCP tunnels let agents reach private MCP servers without exposing them to the public internet.

Self-hosted sandboxes run on customer infrastructure or providers like Cloudflare, Daytona, Modal, and Vercel.

An OpenAI reasoning model improved the best known bound on the planar unit distance problem.

OpenAI published the discrete geometry result as a research milestone for general-purpose reasoning models.

Anthropic acquired Stainless, the SDK and MCP server platform behind every Anthropic SDK to date.

Anthropic detailed months of dialogues with scholars, philosophers, clergy, and ethicists on AI character.

Code with Claude rolled into London with demos and conversations from the teams behind Claude.

Claude Design doubled token limits across every plan.

Cognition's Scott Wu was profiled in The Problem Solvers, with Devin built on Claude.

Genspark's Kay Zhu was profiled in the same series, arguing team quality is the differentiator.