Anthropic Mythos 5 Restored: Who Gets Access (June 27, 2026)
Anthropic Mythos 5 Restored: Who Gets Access (June 27, 2026)
On June 27, 2026, the US government notified Anthropic that Claude Mythos 5 could be redeployed to vetted US organizations responsible for operating and defending critical national infrastructure. The restoration ends a 15-day suspension that began on June 12, 2026, when the government restricted both Mythos 5 and Claude Fable 5 under export-control concerns. The restoration is partial — Fable 5 remains restricted as of June 27 — but represents the first significant unblock of a frontier US AI model under the new staged-release regime. This page covers what restored, who qualifies, what Mythos 5 does, and what comes next.
Last verified: June 27, 2026.
TL;DR
- What restored: Claude Mythos 5 access to vetted US critical-infrastructure organizations only
- What remains restricted: Claude Fable 5 (no public restoration date), general public Mythos 5 access
- Why restored: 15 days of US-government / Anthropic negotiations resolved the national-security review
- Who qualifies: US critical-infrastructure operators, federal agencies, Fortune 500 enterprises with cleared partnerships
- Scale: more than 100 companies and institutions expected in initial restoration wave
- Coincident event: same day as OpenAI’s GPT-5.6 limited-preview launch under similar government controls
- What’s next: Fable 5 restoration negotiations continue; broader Mythos 5 expansion in progress
The 15-day timeline
| Date | Event |
|---|---|
| Pre-June 12 | Claude Mythos 5 and Fable 5 generally available |
| June 9, 2026 | Anthropic releases Claude Fable 5 (1M context, $10/$50 per 1M tokens) |
| June 12, 2026 | US government issues export-control order suspending public access to Mythos 5 and Fable 5 |
| June 12-27, 2026 | Anthropic and federal authorities negotiate resolution |
| June 18, 2026 | Anthropic announces Project Glasswing with ENISA for EU critical-infrastructure access (separate from US suspension) |
| June 26, 2026 | OpenAI previews GPT-5.6 under similar government-controlled access regime |
| June 27, 2026 | US government clears Mythos 5 for vetted US organizations; Anthropic restores access |
| Post-June 27 | Fable 5 restoration negotiations ongoing |
Who actually gets Mythos 5 access
The cleared list is not public, but reporting and the framing (“critical national infrastructure”) indicate the following categories:
Federal government:
- Defense agencies (DoD, defense contractors with CMMC clearance)
- Intelligence community elements
- CISA (Cybersecurity and Infrastructure Security Agency)
- Select FedRAMP-authorized environments
Critical-infrastructure operators:
- Electric utilities and grid operators
- Major financial institutions (banks, exchanges, clearinghouses)
- Telecommunications providers
- Healthcare systems (hospital networks, EHR providers)
- Water and wastewater utilities
- Transportation systems (air, rail, maritime)
Enterprise:
- Fortune 500 companies with existing Anthropic enterprise partnerships
- Major SaaS and cloud providers with critical-infrastructure customer bases
- Cybersecurity vendors and managed security service providers
Research:
- National laboratories with cyber and security research mandates
- Select universities with established cybersecurity research programs and federal contracts
The “more than 100 companies and institutions” figure reported by Japan Times and Engadget gives a rough scale. Expect this list to grow incrementally over the coming weeks as more organizations are cleared.
Who does NOT qualify (as of June 27):
- Individual Claude consumer subscribers (Free, Pro, Max)
- Small and mid-sized businesses without critical-infrastructure designation
- International organizations (EU has a separate Project Glasswing track)
- Most general-purpose enterprise API customers
- Academic researchers without specific cybersecurity-focused grants
What Mythos 5 does that other models don’t
Mythos 5 is Anthropic’s strongest cybersecurity model. Versus Claude Fable 5 or general Claude Sonnet/Opus, Mythos 5 specifically offers:
-
Lifted safeguards in cybersecurity contexts. General Claude models refuse certain offensive-tactics discussions even when the requester is a defender. Mythos 5 is tuned to engage with exploit reasoning, malware behavior analysis, and red-team tactics when the use case is defensive cyber operations.
-
Specialized cybersecurity benchmark performance. Mythos 5 scores 88.0% on Terminal-Bench 2.1 (per OpenAI’s June 26 GPT-5.6 Sol benchmarking) and was the prior frontier on ExploitBench until Sol’s release.
-
Defensive cyber operation workflows. Vulnerability triage, threat detection, incident response acceleration, security report writing, log analysis at scale.
-
Integration with critical-infrastructure security stacks. Anthropic has been building enterprise integrations with major SOC platforms (Splunk, Microsoft Sentinel, Chronicle, etc.) specifically for the critical-infrastructure cleared customer set.
-
High-context security analysis. Mythos 5’s context window supports loading entire codebases, full attack chains, or comprehensive log windows for end-to-end security reasoning.
The Mythos / Sol / Fable competitive landscape after June 27
| Model | Status | Best for | Access |
|---|---|---|---|
| Claude Mythos 5 | Restored to vetted orgs (June 27) | Defensive cyber, vuln research, IR | US critical-infrastructure cleared list |
| GPT-5.6 Sol | Limited preview (June 26) | Frontier coding, security with efficiency | US government cleared partner list |
| Claude Fable 5 | Still restricted as of June 27 | General agent work, 1M context | Awaiting restoration |
| GPT-5.5-Cyber | Generally available | Domain-tuned security workflows | Standard OpenAI API |
| Claude Sonnet 4.x / Opus 4.x | Generally available | General Claude work | Standard Anthropic API |
| Gemini 3.5 Pro | Generally available | Long context, cost | Standard Google API |
What this means for security teams
If you’re on the cleared list: Mythos 5 is back. Resume any paused workflows. Note that Sol may benchmark above Mythos on coding-style security tasks but Mythos 5 retains the lifted-safeguards posture that makes it uniquely suitable for defensive-but-offensively-aware work.
If you’re not on the cleared list: Your alternatives are GPT-5.5-Cyber (Anthropic doesn’t yet have a generally-available “cyber” variant), Claude Sonnet/Opus, or Gemini 3.5 Pro. Build behind a router so you can swap Mythos 5 in when broader access opens.
For all teams: The Mythos 5 / Sol / Fable 5 episodes establish that frontier security-relevant models are now subject to government access controls in the US. Plan for this as the operating environment for the foreseeable future. Build vendor-portable architectures and cultivate eligibility for cleared-partner access if your organization qualifies.
Project Glasswing — the EU parallel track
Separate from the US restoration, Anthropic announced Project Glasswing on June 18, 2026, in partnership with the European Union Agency for Cybersecurity (ENISA), to provide Mythos-class access to EU critical-infrastructure organizations. This is a parallel track that operates under EU regulatory frameworks rather than US export controls. EU organizations should evaluate Project Glasswing rather than waiting for US Mythos 5 access to apply to them.
What to watch
- Fable 5 restoration. No announced date as of June 27. Likely partial restoration mid-July, full public availability in August 2026.
- Mythos 5 access list expansion. The initial 100+ organizations is a starting point. Expect monthly expansion waves.
- GPT-5.6 Sol public availability. Same staged-release pattern. Likely July 2026 ChatGPT Plus/Pro and API GA.
- Equivalent Google process. Gemini 4 will likely go through a similar government-staged review whenever it launches.
- Policy debate intensification. US frontier-AI access controls will face significant industry pushback in Q3 2026.