LLMilestonesA milestone timeline of AI and large language models

Act VI · The frontier and its limits

Closed modelAnthropic

Claude Mythos Preview and Project Glasswing

A model exceptionally good at finding and exploiting software vulnerabilities, offered only to a few organizations through Project Glasswing to harden critical software.

$100Min usage credits for partner organizations

Why it matters

The first frontier model withheld from the public because its cybersecurity abilities were too strong. Seven years earlier GPT-2's release was also delayed over misuse fears; this time the reason was far more concrete.

Costs and controversy

It reportedly found a 27-year-old bug in OpenBSD and a 16-year-old one in FFmpeg. Dual-use capability is itself the risk.

Built on

GPT-2Claude Opus 4.5

Led to

Claude Fable 5

Source: anthropic.com ↗

See it on the 3D timeline →