Anthropic Claude/Mythos/Glasswing/OpenClaw advances, vulns, limits, scaling
Key Questions
What is Anthropic's Mythos Preview?
Mythos Preview is a new AI model from Anthropic, available in private preview on Google Vertex AI. It excels in agentic vulnerability scanning, topping Claude Opus by spotting zero-days and even 27-year-old OpenBSD/Linux exploits in collaboration with partners like Microsoft, AWS, Google, Linux Foundation, Apple, and CrowdStrike.
Why has Anthropic limited the rollout of Mythos?
Anthropic restricted Mythos Preview to a handpicked group due to fears of its powerful hacking capabilities being used for cyberattacks. Despite its strengths in cybersecurity via Project Glasswing, concerns over cyber risks prompted the limited release.
What is Project Glasswing?
Project Glasswing is Anthropic's initiative using Mythos for advanced vulnerability detection, involving partnerships with major tech and cybersecurity firms. It demonstrated spotting long-known exploits, highlighting Mythos's superior agentic scanning over previous models.
What are the reported issues with Claude Code?
Claude Code suffers from persistent memory bugs and regressions, including ignoring configs like CLAUDE.md #44257 and race conditions. These lead to shallow reasoning and declining reliability in complex tasks.
What is Anthropic's current revenue run-rate?
Anthropic announced its run-rate revenue has surpassed $30 billion, up from $9 billion at the end of 2025, driven by high demand for its AI models.
How much compute power is Anthropic scaling to?
Anthropic is scaling to 3.5GW of compute, supported by expanded deals with Google and Broadcom amid skyrocketing demand.
What changes occurred to OpenClaw access in Claude?
As of April 4, 2026, OpenClaw and other third-party tools were removed from Claude subscriptions, requiring payment for access due to outsized risks.
Is there a rival to OpenClaw from China?
China released an open-source framework rivaling OpenClaw, positioning it as a competitive alternative in AI agentic tools.
Mythos Preview limited over cyber/hacking risks despite topping Opus in agentic vuln scanning/zero-days (Glasswing w/MS/AWS/Google/LinuxF/Apple/CrowdStrike); Vertex AI private preview; Claude Code memory bugs/regressions persist; CoPaw OSS framework from China rivals OpenClaw for local agent runs; $30B ARR, 3.5GW compute.