Everything new from OpenAI and Anthropic this week (Week 22, 2026)

OpenAI launched Rosalind Biodefense, updated GPT-5.5 Instant, and brought Codex computer use to Windows, Anthropic released Claude Opus 4.8 and closed a $65B Series H at a $965B valuation, and more

OpenAI launched Rosalind Biodefense and extended trusted access to their GPT-Rosalind life sciences model for select government and allied partners, updated GPT-5.5 Instant for a more natural response style and replaced canvas in GPT-5.5 Instant and Thinking with in-chat writing and code blocks, brought Codex computer use and remote control to Windows, announced plans to sunset GPT-5.2 and GPT-5.3-Codex in Codex and retire o3 and GPT-4.5 from ChatGPT, published their Frontier Governance Framework and a playbook for trustworthy third-party evaluations, detailed their 2026 election safeguards, shared how they built self-improving tax agents with Codex at Thrive Holdings, announced their first Brazil media partnership with Folha de S.Paulo and UOL, added ChatGPT workspace agent controls, a table of contents for long chats, and Workload Identity Federation with an expanded Admin API, released an R&D documentary on their Chip Ganassi Racing collaboration, plus spotted working on a Translation Block widget that supports the fictional High Valyrian language

Anthropic released Claude Opus 4.8 with gains across coding, agentic skills, reasoning, and knowledge work and a standout jump in honesty, raised $65B in Series H funding at a $965B valuation, opened a Milan office, named KiYoung Choi as Representative Director of Korea ahead of a Seoul office opening, shipped a Claude Code security-guidance plugin and a batch of Claude Code reliability fixes, added connector custom roles for Enterprise plans, published an engineering post on how they contain Claude across their products, and had co-founder Chris Olah speak at the Vatican presentation of Pope Leo XIV’s first encyclical on AI, Magnifica humanitas

OpenAI announced their first media partnership in Brazil with Grupo Folha and Grupo UOL, bringing summaries of Folha de S.Paulo and UOL journalism into ChatGPT with attribution and links back to the original sources

At the Vatican, Anthropic co-founder Chris Olah spoke at the presentation of Pope Leo XIV’s first encyclical on AI, Magnifica humanitas, calling for outside critics and moral voices to help steer the technology and noting his team keeps finding mysterious internal structures in the models

[embed]

Anthropic appointed KiYoung Choi, formerly Snowflake’s Korea GM, as Representative Director of Korea ahead of opening a Seoul office, citing Korea as one of the most active markets for Claude.ai

A new Claude Code security-guidance plugin from Anthropic flags and helps fix vulnerabilities as you write code, running via hooks on file edits, after model turns, and on commits, available to all users from the plugin marketplace

[embed]

In an engineering deep-dive, Anthropic detailed how they contain Claude across claude.ai, Claude Code, and Cowork, walking through three isolation patterns and the security failures that taught them to defend the environment layer first

[embed]

OpenAI detailed how they and Thrive Holdings built Tax AI for Crete’s accounting firms using a Codex-driven loop, processing 7,000 returns this season and getting measurably more accurate over time from practitioner corrections

[embed]

In Codex, OpenAI will sunset GPT-5.2 and GPT-5.3-Codex on June 2nd when logged in with a ChatGPT account, with GPT-5.5 becoming the default frontier model on free plans and both models staying available on the API

[embed]

A Claude Code update from Anthropic covers a new full-screen renderer, streamed thinking and tool calls, fewer cryptic errors, compaction progress, more reliable MCP, and self-healing sessions that recover from bad media files

[embed]

OpenAI brought Workload Identity Federation to their API platform so teams can use cloud IAM instead of permanent API keys, and expanded the Admin API with spend alerts, model allowlists, data retention controls, and more granular cost visibility

[embed]

Anthropic is opening a Milan office, their sixth in Europe, to support Italian enterprise, research, and developers, following the release of Pope Leo XIV’s AI encyclical Magnifica humanitas

OpenAI detailed their 2026 election safeguards, including AP live vote counts in the US and Brazil this fall, reliable voting information via Democracy Works, support for cyber defenders, content provenance through SynthID and C2PA, no political ads, and political bias monitoring

[embed]

In R&D Part 1: Here to Win, OpenAI released a documentary on their Chip Ganassi Racing collaboration showing how the IndyCar team uses their tools to analyze race data, sharpen strategy, and optimize pit stops, with footage from Alex Palou’s Long Beach win

[embed]

Anthropic released Claude Opus 4.8, an upgrade over Opus 4.7 with gains across coding, agentic skills, reasoning, and knowledge work at the same price, its standout improvement being honesty, plus effort control in Claude.ai and Cowork, dynamic workflows in Claude Code, and cheaper fast mode

[embed]

OpenAI is updating GPT-5.5 Instant in ChatGPT and the API for a more natural, better-paced response style with fewer overly long or bullet-heavy answers, and is replacing canvas in GPT-5.5 Instant and Thinking with writing and code blocks directly in chat

[embed]

Anthropic raised $65 billion in Series H funding at a $965 billion post-money valuation led by Altimeter, Dragoneer, Greenoaks, and Sequoia, with run-rate revenue crossing $47 billion and recent compute deals with Amazon, Google, Broadcom, and SpaceX

[embed]

OpenAI rolled out new controls for ChatGPT workspace agents in Enterprise and Edu, including GPT-5.5 with reasoning effort, role-based publishing, guided setup, speech output, smarter Slack thread replies, and app templates for GitHub Enterprise, Snowflake, and Databricks

For Enterprise plans, Anthropic added connector permissions to their custom roles, letting admins control which connectors and which individual tools are available to each custom role

OpenAI published their Frontier Governance Framework explaining how their safety and security practices align with emerging legal requirements like California’s Transparency in Frontier AI Act and the EU AI Act’s Code of Practice

OpenAI is retiring o3 from ChatGPT on August 26th and GPT-4.5 on June 27th, with both changes applying only to ChatGPT and not the API

[embed]

OpenAI launched Rosalind Biodefense to help trusted developers build pandemic preparedness tools with GPT-Rosalind, and extended trusted access to the model for select US government and allied public health and biodefense partners

[embed]

OpenAI brought Codex computer use to Windows so it can operate desktop apps by seeing, clicking, and typing, and added remote control of Windows machines from the ChatGPT mobile app

[embed]

A new ChatGPT table of contents from OpenAI organizes conversations with five or more responses, making long chats easier to navigate

[embed]

OpenAI shared a playbook for trustworthy third-party evaluations, arguing that harness choices and validity checks for reward hacking, contamination, and sandbagging are part of the result and not an afterthought

OpenAI were spotted building a Translation Block widget in ChatGPT, with one of the supported languages being High Valyrian, the fictional language from Game of Thrones

[embed]

That brings the week to a close, see you in the next recap. If you want to save time in ChatGPT and Claude, check out AIPRM, our browser extension with over 5,000 prompt templates, Power Continue shortcuts, prompt variables, custom profiles, and a prompt panel for quick access to your favorites, at AIPRM.com.

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