Story perspectives

Anthropic Launches Project Glasswing, Claude Mythos Preview

4/29/2026

Summary

Anthropic Launches Glasswing

Anthropic launched Project Glasswing and previewed Claude Mythos on 2026-04-28.
Over 40 partners such as AWS, Apple, Microsoft, Google, NVIDIA, JPMorgan Stanley and CrowdStrike joined.
Claude Mythos Preview can locate thousands of zero-day flaws and build exploit chains that fully compromise systems.
Anthropic offered up to $100 million in usage credits for the Claude Mythos preview.
Project Glasswing findings will be presented at the 2026 Cyber Summit on May 21.

Summary

White House Eases Ban

The White House is drafting guidance that could let agencies bypass Anthropic’s supply-chain risk label on Mythos AI and overturn an OMB directive barring its use.
On April 28, White House officials met with Anthropic’s CEO.
President Donald Trump told CNBC’s “Squawk Box” a deal with Anthropic is possible.
Sources say the guidance aims to ease Pentagon dispute after it labeled Anthropic a risk due to guardrails on weapons and surveillance.

Summary

Goldman Sachs Bars

Goldman Sachs barred its Hong Kong bankers from accessing Anthropic’s Claude after a strict contract reading.
Gemini and ChatGPT remain available on the bank’s internal AI platform.
Anthropic said Claude has never been officially supported in Hong Kong and declined further comment.
The Hong Kong Monetary Authority noted lenders are assessing extra safeguards for AI-driven cyber risks.

Summary

ECB assesses AI

ECB is contacting eurozone entities to assess preparedness for risks from Anthropic’s Mythos model.
Regulators adopt AI at about half the rate of firms, with only 2 in 10 reporting advanced use.
69 % of respondents rely on OpenAI, 76 % in industry, creating a concentration risk.
Harish Natarajan says only 24 % of authorities collect AI data, 43 % lack plans to start within two years, and regulators lack data and skills.

History

Historical Context

In 2006 the United States and Israel launched Operation Olympic Games, later revealing the Stuxnet worm that in June 2010 sabotaged Iran’s Natanz centrifuges and used four zero-day exploits to spread worldwide. The attack showed how such code can weaponize industrial control systems, prompting regulators to scrutinize Anthropic’s Claude Mythos AI—backed by Microsoft and JPMorgan—as a potential supply-chain cyber threat.

1 of 5

Anthropic Launches Glasswing

Anthropic launched Project Glasswing and previewed Claude Mythos on 2026-04-28.
Over 40 partners such as AWS, Apple, Microsoft, Google, NVIDIA, JPMorgan Stanley and CrowdStrike joined.
Claude Mythos Preview can locate thousands of zero-day flaws and build exploit chains that fully compromise systems.
Anthropic offered up to $100 million in usage credits for the Claude Mythos preview.
Project Glasswing findings will be presented at the 2026 Cyber Summit on May 21.

1 / 5