18 days that changed the industry

On June 9, 2026, Anthropic launched Claude Fable 5 and Claude Mythos 5 — its most capable models ever publicly available. Three days later, on June 12, the US government applied export controls to both, requiring Anthropic to restrict access to foreign nationals inside and outside the US. Unable to verify nationality in real time, Anthropic suspended both models for all users.

On June 30, Commerce Secretary Howard Lutnick sent a letter withdrawing the export-control license requirement. On July 1, 2026, Claude Fable 5 returned globally. Mythos 5 had access restored only for authorized US organizations via Project Glasswing.

What triggered the block

The trigger was an Amazon researchers report identifying a bypass method for Fable 5 safeguards — prompts that made the model identify software vulnerabilities. In one case, it produced code demonstrating how to exploit a specific vulnerability.

Anthropic responded with its own testing and found something revealing: the behavior was not unique to Fable 5. Less capable models — including Claude Opus 4.8, GPT-5.5, and Kimi K2.7 — identified the same vulnerabilities. For the exploitation demonstration, every model tested produced the same result as Fable 5. Anthropic argued the case was "borderline" for Fable safeguards and exposed no unique Mythos-level capabilities.

What returned — and what changed

Fable 5 returned with a new safety classifier trained to block the specific reported behavior in over 99% of cases. Blocked requests are not outright refused — they are routed to Claude Opus 4.8, with user notification. The tradeoff: more false positives in routine coding and debugging tasks.

At launch: 1 million token context window, up to 128K output tokens, $10/M input and $50/M output. For Pro, Max, Team, and select Enterprise plans, access was included for up to 50% of weekly limits through July 7 — later extended to July 12 as a competitive response to GPT-5.6 launch.

The precedent that remains

The episode inaugurated a new era: frontier model launches stopped being product releases and became negotiated deployments shaped by US national security review. Companies that built critical workflows on Fable 5 had no alternative for 18 days. The market responded by moving toward model-agnostic fallback architectures — proxy layers that route to local open-weight models when proprietary APIs go offline. Anthropic stated it is deepening cooperation with the government on pre-release testing, information sharing, and research collaboration.

Primary sources:
- Anthropic — Redeploying Fable 5: anthropic.com
- VentureBeat: venturebeat.com
- MacRumors: macrumors.com
- Neowin: neowin.net