GPT-56T 861 —
MUSE-SPK 835 -0.7%
GPT-56SC 828 -5.2%
QWEN-38X 824 —
CL-OP55X 822 —
GROK-46H 822 -5%
GPT-6A 820 —
GLM-5 784 -8.4%
CL-FAB5H 743 -5.6%
KIMI-K3X 742 -8.4%
CL-OP5H 720 -5.8%
CL-OP5X 709 -18%
CL-OP46H 698 -5.9%
CL-OP47H 690 -5.9%
GEM-38FH 677 +0.1%
GEM-37FH 657 -24%
GPT-56S 622 —
CL-OP47 582 -0.7%
GPT-55H 582 —
INKL 531 —
GEM-31P 513 —
GEM-3P 499 —
CL-OP46 496 -0.2%
CL-OP48 490 —
GPT-56T 861 —
MUSE-SPK 835 -0.7%
GPT-56SC 828 -5.2%
QWEN-38X 824 —
CL-OP55X 822 —
GROK-46H 822 -5%
GPT-6A 820 —
GLM-5 784 -8.4%
CL-FAB5H 743 -5.6%
KIMI-K3X 742 -8.4%
CL-OP5H 720 -5.8%
CL-OP5X 709 -18%
CL-OP46H 698 -5.9%
CL-OP47H 690 -5.9%
GEM-38FH 677 +0.1%
GEM-37FH 657 -24%
GPT-56S 622 —
CL-OP47 582 -0.7%
GPT-55H 582 —
INKL 531 —
GEM-31P 513 —
GEM-3P 499 —
CL-OP46 496 -0.2%
CL-OP48 490 —
← Back to feed

Anthropic Embeds Mythos-1 in Claude Code Source — Late-June Commercial Window Exposed

Anthropic is preparing Mythos-1 for its first commercial deployment inside Claude Code and Claude Security. Source code analysis published by TestingCatalog on May 23 found new strings in the Claude Code client that were absent until recently, including an explicit reference: “Access to the Claude Mythos model in Claude Code and Claude Security.” The internal API tag is claude-mythos-1-preview, following the same naming convention Anthropic uses before general release.

Users briefly spotted a model labeled “Mythos 1” inside the Claude Code interface before the selection disappeared again. Separate traces of the model have also surfaced in AWS and Google Cloud vulnerability discovery systems without the Preview tag — a pattern Anthropic’s internal teams have removed each time.

What the Numbers Actually Are

The stakes behind the rollout are not theoretical. On Anthropic’s internal Firefox benchmark — vulnerabilities from Firefox 147’s JavaScript engine, all patched in Firefox 148 — Opus 4.6 produced 2 working exploits. Mythos Preview produced 181 working exploits and achieved register control in 29 additional cases.

On OSS-Fuzz with 7,000 entry points, Opus 4.6 and Sonnet produced 250-275 crashes at tier 1-2, zero at tier 4-5. Mythos Preview produced 595 at tier 1-2 and 10 complete control-flow hijacks at tier 5 on fully patched targets. Human validation put 89% of findings as accurate and 98% within one severity level of the declared rating.

One documented case: Mythos Preview autonomously identified a 17-year-old remote code execution in the FreeBSD NFS daemon (CVE-2026-4747), built a 20-gadget ROP chain across 6 sequential RPC packets, exploited a 304-byte buffer overflow, and wrote its own SSH key into root’s authorized_keys without human intervention beyond the initial request.

The Glasswing Bottleneck

Glasswing, Anthropic’s restricted partner programme, has now disclosed 1,596 vulnerabilities across 281 open-source projects, with 97 patched upstream as of May 22. Finding pace has outrun fix pace by roughly 16 to 1. The integration of Mythos-1 into Claude Code is partly designed to address this: surfacing vulnerabilities closer to the moment of code creation, tied to the pull request and the engineer who can act on it, rather than landing in a separate queue.

Anthropic updated Glasswing’s framing on May 22, noting the model is now helping protect open-source projects beyond its original partner set and stating: “In the near future, once we’ve developed the far stronger safeguards we need, we look forward to making Mythos-class models available through a general release.”

What the Timeline Looks Like

Access will remain filtered through Anthropic’s vetting when Mythos-1 launches. The expected pattern: Enterprise Claude Code users with Glasswing or Cyber Verification Program status first, paired with the release of Opus 4.8 (or whatever commercial name it takes). Multiple observers watching Anthropic’s product cadence put the window at late June to early July 2026.

The permission framework and dedicated UIs are already in the code. The delay is not engineering — it is the safeguard layer Anthropic says it needs before any model at Mythos-class capability goes broader. Current Claude Code users will not see the model appear in the selector overnight.

Key Numbers

MetricOpus 4.6Mythos-1
Firefox 147 exploits (working)2181
OSS-Fuzz tier 5 control-flow hijacks010
Human validation accuracy—89%
Glasswing disclosures to date—1,596