# Four AI Labs Split on Model Access This Week > Five stories from the week of April 7-14, 2026, covering how Anthropic, OpenAI, Meta, and Google each chose fundamentally different AI model access strategies — from restricted security consortiums to full Apache 2.0 open source. Content type: article Source URL: https://www.agentpmt.com/articles/four-ai-labs-split-on-model-access-this-week Markdown URL: https://www.agentpmt.com/articles/four-ai-labs-split-on-model-access-this-week?format=agent-md Updated: 2026-04-15T05:32:45.124Z Author: Stephanie Goodman Tags: AI Powered Infrastructure, Security In AI Systems, News --- # Four AI Labs Split on Model Access This Week The week of April 7-14, 2026, produced four fundamentally different answers to the same question: who gets access to the most powerful AI models? Here are the five stories that defined the split. * * * ## Anthropic Restricts Mythos to a Security Consortium Anthropic announced that Claude Mythos Preview — its most advanced unreleased model — would be available only to roughly 50 organizations through Project Glasswing. The decision followed weeks of internal testing where Mythos scanned major operating systems and critical software, identifying thousands of unknown vulnerabilities and creating working exploits in 83.1 percent of cases. AWS, Apple, Google, Microsoft, JPMorganChase, CrowdStrike, and the Linux Foundation joined as launch partners. The [AI cybersecurity](https://www.agentpmt.com/articles/why-ai-cybersecurity-needs-quantum-safe-signatures-now) implications drove regulators to act — Treasury Secretary Bessent and Fed Chair Powell convened Wall Street CEOs to discuss the findings. **Source:** Fortune * * * ## OpenAI Launches Identity-Based Tiered Access OpenAI took the inverse approach with GPT-5.4-Cyber, expanding its Trusted Access for Cyber program. Rather than restricting which model users can access, OpenAI verifies the user and unlocks capabilities accordingly. Verified cybersecurity professionals get fewer restrictions on vulnerability research and analysis. The bet: broad access with strong identity controls is both safer and more commercially viable than locking models behind a consortium. **Source:** Axios * * * ## Meta Ships Its First Proprietary Model Meta launched Muse Spark on April 8 — its first AI model released without open weights, breaking from the Llama open-source strategy that had defined its approach since 2023. CEO Mark Zuckerberg reportedly grew dissatisfied with Llama's competitive position relative to ChatGPT and Claude, driving the pivot. Meta plans to deploy Muse Spark across Facebook, Instagram, WhatsApp, and Ray-Ban Meta AI glasses, with a paid API preview for select developers. **Source:** 24/7 Wall St * * * ## Google Goes Full Open Source With Gemma 4 Google released Gemma 4 under Apache 2.0 — the most permissive license in Gemma history, removing every restriction that blocked enterprise adoption. Performance jumped dramatically: AIME 2026 math scores rose from 20.8 to 89.2 percent compared to Gemma 3. The strategic logic is transparent — if the most capable open-source models carry Google's DNA, Google Cloud becomes the natural scaling destination when companies move from prototyping to production. **Source:** Google Open Source Blog * * * ## The Market Supports All Four Approaches A Research and Markets report projects the open-source AI model market will reach $50 billion by 2030. The fragmentation visible this week is likely the new normal — the market is large enough that restricted consortiums, tiered access, proprietary releases, and full open source can coexist. For AI agents and the [AI agent platform](https://www.agentpmt.com/articles/why-agent-marketplaces-matter-the-economics-of-not-reinventing-the-wheel) teams building on them, model-agnostic infrastructure becomes more valuable as access strategies diverge. **Source:** GlobeNewsWire / Research and Markets * * * ## Sources - Anthropic's Mythos reveals a growing security gap — Fortune - OpenAI rolls out tiered access to advanced AI cyber models — Axios - Meta Platforms Finally Releases Muse Spark — 24/7 Wall St - Gemma 4: Expanding the Gemmaverse with Apache 2.0 — Google Open Source Blog - Open-Source AI Model Market Research Report 2026 — GlobeNewsWire / Research and Markets