Your OpenAI and Anthropic weekly digest (Week 16, 2026)

OpenAI announced Codex update with background computer use, GPT-Rosalind life sciences model, Cloudflare Agent Cloud partnership, Anthropic released Claude Opus 4.7 and Claude Design, routines in Claude Code research preview, and more

OpenAI announced Cloudflare Agent Cloud partnership with OpenAI frontier models including GPT-5.4, Agents SDK update with native sandbox execution and model-native harness, Codex update with background computer use, image generation, memory, automations, and 90+ new plugins, introduced GPT-Rosalind life sciences reasoning model in research preview, Trusted Access for Cyber partners with GPT-5.4-Cyber, shared ChatGPT tax queries data, life sciences usage report, and closed gender gap data, re-enabled Study mode in ChatGPT, ChatGPT keyboard shortcut customization, ads rollout on Free and Go plans in Australia, New Zealand, and Canada, TeenAegis AI Danger Index lowest risk score, and Kevin Weil (OpenAI for Science lead and former CPO), Srinivas Narayanan (CTO of enterprise applications), and Bill Peebles (head of Sora) departures, with OpenAI for Science decentralized into other research teams and Prism sunset folded into Codex

Anthropic released Claude Opus 4.7, launched Claude Design in research preview by Anthropic Labs, introduced routines in Claude Code research preview for scheduled, API, and webhook-triggered automations, redesigned Claude Code desktop app for parallel agents, appointed Vas Narasimhan to Board of Directors by the Long-Term Benefit Trust, published Automated Alignment Researchers study on weak-to-strong supervision, and made Claude for Word available on Pro and Max plans

Cloudflare and OpenAI expanded their partnership making OpenAI frontier models including GPT-5.4 available to millions of Agent Cloud customers, with Codex harness generally available in Cloudflare Sandboxes

[embed]

Anthropic’s Long-Term Benefit Trust appointed Novartis CEO Vas Narasimhan to the Board of Directors, making Trust-appointed directors a majority of the Board

[embed]

Anthropic Fellows study ran nine Claude Opus 4.6 copies as Automated Alignment Researchers on weak-to-strong supervision, closing 97% of the performance gap versus 23% human baseline over 800 hours at $18k in tokens and training expenses

[embed]

Anthropic introduced routines in Claude Code research preview that run on their web infrastructure via schedule, API endpoint, or GitHub webhook triggers, available on all paid plans with Claude Code on the web

[embed]

OpenAI shared that US ChatGPT tax-related queries grew 4x from Q1 2025 to Q1 2026, with roughly a third on earnings and withholding and over 30% on filing forms and tax software

[embed]

Anthropic redesigned the Claude Code desktop app for parallel agents with a session sidebar, integrated terminal, in-app file editor, faster diff viewer, HTML/PDF preview, and drag-and-drop layout with CLI plugin parity

[embed]

OpenAI released an updated Agents SDK with a model-native harness, native sandbox execution, and Manifest abstraction separating harness from compute for security and durability, with support for Blaxel, Cloudflare, Daytona, E2B, Modal, Runloop, and Vercel

[embed]

OpenAI shared that the roughly 80% male first-name skew ChatGPT had at launch is now gone according to their anonymized data

[embed]

OpenAI published a life sciences report estimating 200,000 weekly global users do advanced life sciences work in ChatGPT, with around 75% on literature review, plus customer stories from Valius Sciences, Sequencing.com, Junevity, Ginkgo Bioworks, and Retro Biosciences

[embed]

OpenAI shipped a major Codex update with background computer use on macOS, in-app browser, gpt-image-1.5 image generation, memory preview, scheduled automations that can work across days or weeks, and 90+ new plugins

[embed]

OpenAI introduced GPT-Rosalind, the first in a new Life Sciences frontier reasoning model series, available in research preview via Trusted Access to qualified Enterprise customers, with a freely accessible Life Sciences research plugin for Codex

[embed]

Anthropic released Claude Opus 4.7 generally available across all Claude products and the API at $5/$25 per million tokens, with improved coding, higher-resolution vision, a new xhigh effort level, cyber safeguards, and new /ultrareview slash command in Claude Code

[embed]

OpenAI shared Trusted Access for Cyber partners including Bank of America, BNY, Cisco, CrowdStrike, NVIDIA, Oracle, Palo Alto Networks, and Zscaler, committed $10M in API credits via the Cybersecurity Grant Program with initial recipients Socket, Semgrep, Calif, and Trail of Bits, and provided GPT-5.4-Cyber to CAISI and UK AISI for evaluations

[embed]

OpenAI began rolling out ads for Free and Go plan users in Australia, New Zealand, and Canada, with Plus, Pro, Business, Enterprise, and Education plans remaining ad-free

OpenAI reversed the silent removal of Study mode in ChatGPT after user feedback, bringing it back behind the ‘/’ menu until something better is in place

[embed]

Anthropic Labs launched Claude Design in research preview powered by Claude Opus 4.7 for collaborative visual work like prototypes, slides, and mockups with team design system support and Claude Code handoff, available for Pro, Max, Team, and Enterprise subscribers

[embed]

ChatGPT added full keyboard shortcut customization, remapping, and disabling via Help > Keyboard Shortcuts or cmd+/ as a major accessibility improvement

[embed]

Anthropic made Claude for Word available on Pro and Max plans to use alongside Claude Opus 4.7

[embed]

Kevin Weil (OpenAI for Science lead and former CPO), Srinivas Narayanan (CTO of enterprise applications), and Bill Peebles (head of Sora) are leaving OpenAI, with OpenAI for Science decentralized into other research teams and Prism sunset with its team folded into Codex under Thibault Sottiaux

[embed]

OpenAI received the lowest risk score among evaluated providers in TeenAegis’ inaugural AI Danger Index

[embed]

That’s what happened this week — see you next weekend!

Follow me on X/LinkedIn/Threads/Bluesky for more