OpenAI and Anthropic week in review (Week 15, 2026)

OpenAI introduced $100 Pro tier and acquired Cirrus Labs, Anthropic revealed $30B run rate and launched Project Glasswing with Claude Mythos Preview, Managed Agents, and Cowork GA, and more

Here’s what OpenAI and Anthropic have been up to this past week

OpenAI published industrial policy ideas for the Intelligence Age, announced Safety Fellowship pilot program, acquired Cirrus Labs for agent infrastructure, introduced new $100 Pro tier with 5x Codex usage over Plus as the existing $200 Pro tier remains the highest option, published child safety blueprint, announced over $100 million in OpenAI Foundation grants for Alzheimer’s research, shared enterprise AI update with enterprise now over 40% of revenue, released GPT-5.3 Instant Mini as new fallback model, announced older Codex model retirements, added Outlook shared mailboxes and calendars support in ChatGPT, paused UK Stargate project over energy costs and regulation, disclosed macOS app signing security incident from Axios supply chain compromise, and launched Prism Paper Review, plus spotted ImageGen 2 A/B testing in ChatGPT and European Commission plans to designate ChatGPT as very large online search engine under the Digital Services Act

Anthropic expanded partnership with Google and Broadcom for multiple gigawatts of next-generation compute with run rate surpassing $30B and over 1,000 enterprise customers now spending $1M+ annually, officially announced previously leaked Claude Mythos Preview through Project Glasswing finding thousands of zero-day vulnerabilities with $100M in usage credits, launched Managed Agents in public beta and published engineering blog on decoupling agent architecture, published trustworthy agents framework update, made Claude Cowork generally available on all paid plans with enterprise controls, introduced the advisor tool for pairing Opus with Sonnet or Haiku, and launched Claude for Word beta, and more

OpenAI published exploratory policy ideas for the Intelligence Age with fellowships and research grants of up to $100,000 and $1 million in API credits, plus a new OpenAI Workshop opening in May in Washington, DC

[embed]

OpenAI announced the Safety Fellowship, a pilot program for external researchers running September 2026 through February 2027 with stipend, compute support, and mentorship at Constellation in Berkeley

[embed]

Anthropic expanded its compute partnership with Google and Broadcom for multiple gigawatts of next-generation TPU capacity starting in 2027, with run rate surpassing $30 billion and enterprise customers spending $1M+ annually doubling from 500 to over 1,000 in under two months

[embed]

Cirrus Labs announced it is joining OpenAI as part of the Agent Infrastructure team, with source-available tools including Tart being relicensed more permissively and Cirrus CI shutting down June 1, 2026

[embed]

OpenAI is retiring older Codex models on April 14 when signing in with a ChatGPT account, keeping gpt-5.4, gpt-5.4-mini, gpt-5.3-codex, and gpt-5.3-codex-spark as supported models

[embed]

OpenAI’s Kevin Weil announced Paper Review in Prism, an AI workflow for reviewing technical and scientific papers that checks math, derivations, structure, and whether claims are supported by results

[embed]

Anthropic announced Project Glasswing with AWS, Apple, Broadcom, Cisco, CrowdStrike, Google, JPMorganChase, the Linux Foundation, Microsoft, NVIDIA, and Palo Alto Networks using the unreleased Claude Mythos Preview that autonomously found thousands of zero-day vulnerabilities in every major OS and browser, with $100M in usage credits and $4M in open-source security donations, confirming the model previously leaked through publicly accessible draft posts

[embed]

Anthropic published an engineering blog on the Managed Agents architecture explaining how they decoupled the “brain” from the “hands” and the session, reducing p50 time-to-first-token by roughly 60% and p95 by over 90%

[embed]

OpenAI Foundation announced over $100 million in grants across six research institutions to accelerate Alzheimer’s research through AI-assisted causal mapping, drug design, biomarker discovery, and testing of off-patent treatments

[embed]

OpenAI introduced the Child Safety Blueprint, a policy framework developed with NCMEC, the Attorney General Alliance, and Thorn for modernizing laws around AI-generated CSAM, improving reporting, and building safety-by-design measures

[embed]

Anthropic launched Claude Managed Agents in public beta with secure sandboxing, long-running sessions, multi-agent coordination in research preview, and scoped permissions, with Notion, Rakuten, Asana, Vibecode, and Sentry already building on it

[embed]

ChatGPT Outlook Email and Calendar apps now support shared mailboxes and calendars including reading messages, sending email from shared mailboxes, and managing events on shared calendars

OpenAI CRO Denise Dresser shared an enterprise update noting enterprise now makes up over 40% of revenue on track to reach parity with consumer by end of 2026, Codex at 3 million weekly active users, and APIs processing over 15 billion tokens per minute

[embed]

Anthropic published a practical update on their trustworthy agents framework covering human control, goal alignment, and prompt injection defense in specific product decisions, with calls for industry-wide benchmarks and open standards

Anthropic made Claude Cowork generally available on all paid plans with enterprise controls including role-based access, group spend limits, usage analytics, expanded OpenTelemetry, Zoom MCP connector, and per-tool connector controls

[embed]

OpenAI introduced a new $100 per month Pro tier with 5x Codex usage over Plus between the $20 Plus and existing $200 Pro which remains the highest option at 20x, while the Plus Codex promo ended and usage was rebalanced for more sessions throughout the week

[embed]

Anthropic introduced the advisor tool letting developers pair Opus as advisor with Sonnet or Haiku as executor, where the smaller model drives the task and escalates only when needed, reducing cost per task while approaching Opus-level performance

[embed]

OpenAI’s UK Stargate project has been paused due to the cost of energy, delays in accessing the national grid, and the UK regulatory environment including copyright rules

[embed]

OpenAI released GPT-5.3 Instant Mini in ChatGPT replacing GPT-5 Instant Mini as the fallback model users reach after hitting rate limits, with stronger writing and contextual awareness

Anthropic launched Claude for Word in beta on Team and Enterprise plans with sidebar editing, tracked changes, comment-based editing, and shared context across Claude for Excel and PowerPoint

[embed]

According to Handelsblatt, the European Commission plans to designate ChatGPT as a very large online search engine under the Digital Services Act, putting it in the same regulatory category as Google Search

[embed]

OpenAI disclosed a macOS app signing security incident from a compromised Axios library in a GitHub Actions workflow, with signing certificates being rotated and older app versions losing support after May 8, though no evidence of compromise was found

[embed]

Spotted OpenAI testing ImageGen 2 in ChatGPT through an A/B test with one of the images generated using the “imagegen2-dust-dec-ev3” variant

[embed]

That’s it for this week — see you next weekend!

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