Sources: about 200 companies accessing Mythos Preview through Project Glasswing have preserved their access despite a recent US government shutdown order
Context & Ripple Effects
Project Glasswing had been expanding Mythos Preview access beyond its initial circle, with reported outreach to allied governments, major companies, and potentially EU cyber authorities. At the same time, reported use within US national-security agencies and a jailbreak-linked restriction debate put the model’s cyber capabilities under unusually close scrutiny.
The reported continued access for a defined corporate group matters because it suggests the shutdown order has not produced a uniform halt across the program’s existing users.
First-order effects
- About 200 companies already using Mythos Preview through Project Glasswing can continue working with the model despite the shutdown order.
- Anthropic retains an active enterprise cohort for Mythos Preview while the government’s restrictions are being applied or interpreted.
Second-order effects
- Organizations with continuing access may gain an operational advantage in software and cyber-related workflows over firms excluded from the preview, increasing pressure for clearer eligibility rules.
- The mismatch between a shutdown order and preserved access will intensify scrutiny of the controls, risk assessments, and contractual boundaries governing Project Glasswing users.
Third-order effects
- If selective access persists, frontier models with demonstrated cyber capability may be governed through segmented, trusted-user distribution rather than broad availability or outright prohibition.
- The episode points to a durable tension in AI policy: governments and enterprises want the defensive and productivity benefits of advanced models while trying to limit their offensive misuse potential.
The trend: Frontier AI deployment is moving toward access regimes that differentiate among users and jurisdictions based on perceived cyber risk and institutional trust.