Claude Security (Mythos 5 scans)
Claude Security (Mythos 5 scans)
Vintage: 2026-08. Primary evidence is an official @claudeai X thread dated 2026-08-21. Not @AnthropicAI and not an Anthropic blog. Linked URL in the first post was not fetched. The last post is truncated in the API at "expanding our Cyber" — do not invent the rest of that sentence.
One-line summary: The official Claude product account said Claude Security scans now run on Claude Mythos 5, in public beta for all Claude Enterprise customers, with no separate model access; findings include CWE / confidence / severity / a suggested fix; patches open in Claude Code on the web; Defender Advantage Fund provides $35M in credits for open-source security.
What it is
A Claude product surface named "Claude Security." The 2026-08-21 @claudeai thread states that scans now run on Claude Mythos 5, available that day in public beta for all Claude Enterprise customers, and that the model runs behind the scan (findings only — "no separate model access needed"). Treat this as a product-account X thread, not an @AnthropicAI lab announcement and not a rewrite of the June government release gate on government-gated-frontier-releases.
Why it matters to this thread
AI-as-a-tool-layer and AI safety / security deployment are in-scope. This is the first dated official @claudeai claim the thread has that Mythos 5 is the scan engine for a named security product, with Enterprise public-beta availability and an explicit "model stays behind the scan" access model. Distinct from vibe-coding-security (Escape.tech scan of shipped vibe-coded apps — different method, different sample).
Key facts (from 2026-08-25-x-ai-news-25-aug-2026-openai-sol-price-cut-claude-security)
All bullets are X posts (fetch_method: x-mcp). Permalinks live on the source page.
- Availability (X post, @claudeai, 2026-08-21): "Claude Security scans now run on Claude Mythos 5, available today in public beta for all Claude Enterprise customers." "Put our most capable security model to work on your codebase, no separate model access needed." Linked
t.co/zJSUgGjtZFwas not fetched. - Scan behavior (same-thread reply): "Point Claude Security at a GitHub repo and Mythos scans for vulnerabilities tracing data across files and reasoning about how components interact. Each finding comes back with a CWE category, confidence and severity ratings, and a suggested fix."
- Patches / access model (same-thread reply): "Suggested patches open in Claude Code on the web, using the models your team already uses." "These updates are part of our work to give defenders greater access to Mythos results without requiring direct access to the model: the model runs behind the scan and returns findings only."
- Defender Advantage Fund (same-thread reply): "It's one of several steps to bring frontier capability to more defenders: we’re working with partners to integrate Mythos 5 into their security products and services, our new Defender Advantage Fund provides $35M in credits for open-source security, and we’re expanding our Cyber" — truncated in the API at "expanding our Cyber." Do not invent the rest of the sentence. The $35M figure is in this hydrated post (credits for open-source security).
What this source does not establish
- Not an @AnthropicAI lab post and not a fetched Anthropic blog. @AnthropicAI had no original posts in the fetch window.
- Enterprise public beta only, as written. No Team / Pro / free-tier claim.
- No direct Mythos 5 model access. The thread says the model runs behind the scan and returns findings only.
- No vulnerability counts, precision/recall, or benchmark. CWE / confidence / severity / suggested fix are the output fields as written.
- Not a government-gate change. Do not collapse into government-gated-frontier-releases (June 2026 Mythos 5 ~100-company gate). Enterprise scan access with the model behind the scan is not the same claim as ungated direct Mythos 5. Left open — not adjudicated.
- Do not complete the truncated "expanding our Cyber" sentence.
Adjacent 5.1 release (not a scan-engine rewrite)
- From 2026-09-02-grok-com-ai-news-digest-2026-09-02-fable-5-1-astra-atlas-g20 (Anthropic announcement, September 1, 2026): Mythos 5.1 is a later point update, the same model as Fable 5.1 with different safeguards, currently available to a set of US organizations through trusted access programs; Cyber Verification Program Mythos-class access is "in the near future." Do not collapse that into this page. This page remains Mythos 5 behind Enterprise scans (August 21 @claudeai). Full treatment: claude-fable-5-1.
- From 2026-09-02-x-ai-overnight-fable-5-1-astra-critical-reward-seeker (official @claudeai X, 2026-09-01): Mythos 5.1 "stays in trusted-access programs for cyberdefenders and life scientists." Same adjacent-release note — not a scan-engine rewrite.
- From 2026-09-10-overnight-x-openai-defense-factory-uk-aisi-mythos-51-access: OpenAI openai-defense-factory is a different lab’s playbook; UK AISI Mythos 5.1 access reportage lives on uk-aisi-mythos-51-access. Neither is a Claude Security scan-engine rewrite. This page remains Mythos 5 behind Enterprise scans.
Open questions
- Does "all Claude Enterprise customers" include every Enterprise SKU from day one, and in which surfaces (claude.ai vs claude-code vs a separate Security UI)?
- What does the unfetched linked page add (docs, pricing, data-retention) that the X thread omits?
- Does the June customer-by-customer direct-model gate still apply, given this thread's "no separate model access" / "returns findings only" language? This source does not say.
Sources
- 2026-08-25-x-ai-news-25-aug-2026-openai-sol-price-cut-claude-security
- 2026-09-02-x-ai-overnight-fable-5-1-astra-critical-reward-seeker — adjacent Mythos 5.1 trusted-access; not a scan-engine rewrite
- 2026-09-10-overnight-x-openai-defense-factory-uk-aisi-mythos-51-access — Defense Factory / AISI access are adjacent, not a scan-engine rewrite