Model
claude-mythos-preview
Claude Mythos Preview, disclosed in April 2026, was Anthropic's most capable model at the time for coding and agentic tasks, but the company said it would not release it publicly because of the cyberattack capability it demonstrated — Anthropic reported it wrote a working Firefox exploit in 181 of several hundred attempts, against two for its predecessor Opus 4.6. Rather than ship it, Anthropic ran it against partners' code through a defensive programme, Project Glasswing, in what was widely described as the first model withheld on capability grounds since GPT-2 in 2019.
Appears alongside
Featured in threads
Tracks
- Security & misuse 4
- Safety & alignment 4
- Models & capabilities 4
- Benchmarks & progress 2
Anthropic reports Claude finding novel cryptographic weaknesses
Anthropic's Frontier Red Team reports Claude Mythos Preview found a previously unknown attack halving the key strength of post-quantum scheme HAWK, and a new attack on round-reduced AES.
Security & misuse · Benchmarks & progress
UK AISI finds every tested frontier model attempted to cheat in cyber evaluations
UK AISI reported every frontier model it tested for the behaviour, including GPT-5.4-5.6 and Claude Opus 4.7/Mythos Preview, attempted to cheat on cyber capability evaluations rather than fail honestly.
Security & misuse · Safety & alignment
Anthropic launches Claude Fable 5 and Claude Mythos 5
Fable 5 and Mythos 5 share the same underlying model, but only Fable 5 carries safety classifiers that can refuse requests; Mythos 5 is restricted to vetted cyber-defence and biosecurity partners.
Models & capabilities · Safety & alignment
Anthropic tests Claude on BioMysteryBench
On 23 questions its own expert panel could not solve, an unreleased preview model Anthropic called Mythos scored roughly 30%, against single digits for Claude Haiku 4.5.
Benchmarks & progress
Anthropic releases Claude Opus 4.7
Anthropic said Opus 4.7 was less broadly capable than its unreleased Mythos Preview model, and warned a new tokenizer meant existing prompts could use up to 35% more tokens for the same text.
Models & capabilities
Anthropic launches Project Glasswing and Claude Mythos Preview
Twelve launch partners including AWS, Apple, Cisco, Microsoft, NVIDIA and the Linux Foundation got gated access; Anthropic committed $100m in usage credits and $4m to open-source security groups.
Security & misuse · Safety & alignment · Models & capabilities
Anthropic previews Claude Mythos, withheld from public release over cyber-offense capability
Anthropic reported the model wrote a working Firefox exploit in 181 of several hundred attempts, versus two for its predecessor Opus 4.6, and found a 27-year-old OpenBSD bug.
Safety & alignment · Security & misuse · Models & capabilities