Economy June 19, 2026 01:24 AM

Some Early Anthropic Mythos Testers Still Have Preview Access After U.S. Block

Selected Glasswing participants retained entry to Mythos Preview even as other versions were shut down under national security export controls

By Marcus Reed
Share
Twitter Reddit Facebook LinkedIn

Anthropic has kept access to its Mythos Preview open for a subset of early testers within its Glasswing program, despite a U.S. government directive that resulted in the shutdown of other releases. The preview, limited to roughly 200 organizations after the model flagged thousands of software vulnerabilities, remains available to certain firms while broader access for foreign nationals was suspended on national security grounds.

Some Early Anthropic Mythos Testers Still Have Preview Access After U.S. Block
Summarize with
ChatGPT Perplexity Claude Grok Gemini

Key Points

  • Anthropic reserved Mythos Preview access for roughly 200 organizations under its Glasswing program after the model identified thousands of software vulnerabilities - impacts technology and cybersecurity sectors.
  • A publicly released, less powerful version of Mythos was disabled following U.S. export control directives that suspended access for all foreign nationals citing national security concerns - impacts global AI deployment and export-sensitive tech markets.
  • Some early Glasswing participants, including Dragos and Cisco Systems, confirmed they retained access to the Mythos Preview while others were denied - affects cybersecurity vendors and government-corporate relations.

June 18 - Anthropic has continued to allow some of the organizations it picked early to test the Mythos AI system to use a preview build, even as a U.S. government order prompted the complete shutdown of other versions of the model. The preview environment, provided under Anthropic's Glasswing program, was restricted to about 200 organizations after Mythos flagged thousands of software vulnerabilities.

Anthropic initially released a less capable public variant of Mythos, but that version was later disabled in response to Washington's export control directives, which suspended access to the model for all foreign nationals on national security grounds. Despite that suspension, several firms that had been part of the early Glasswing test cohort have retained their preview access.

Security-focused companies Dragos and Cisco Systems confirmed they continue to have access to the Mythos Preview. It remains unclear how Anthropic is deciding which Glasswing members keep access and which do not.

Company representatives at Anthropic, Cisco and Dragos were not immediately available for comment outside business hours. Similarly, the European cybersecurity agency Enisa, which had been invited to join Glasswing before the U.S. government order was issued, was told on Friday that it would no longer be granted access. Enisa was also not immediately available for comment outside business hours.

The episode follows a widening rift between Anthropic and the U.S. government earlier this year, when the company declined requests to permit the U.S. military to use its AI models for domestic surveillance and for fully autonomous weapons systems. That breakdown in relations is part of the broader context surrounding the decision to restrict certain Mythos versions.

The selective continuation of Mythos Preview access raises questions among industry observers about consistency of enforcement and about the mechanics Anthropic is using to vet and authorize individual members of its Glasswing program. For now, the company appears to be maintaining a limited, invite-only preview for some partners while complying with directives that remove broader foreign-national access to publicly released builds.


Market note: The report included mention of Cisco Systems alongside Anthropic; Cisco's quoted price movements and related market commentary were provided as part of the original coverage.

Risks

  • National security-driven restrictions can produce uneven access to advanced AI models, creating uncertainty for international partners and vendors in the tech and cybersecurity sectors.
  • Lack of transparency about how Anthropic selects which Glasswing members retain access could undermine trust among invited organizations and complicate collaboration in cybersecurity testing.
  • Ongoing tensions between Anthropic and the U.S. government over military use of AI models may influence future procurement, partnership decisions, and regulatory scrutiny in defense and AI industries.

More from Economy

ECB’s Wunsch Says Another Rate Hike Remains an Option as Iran Deal Lowers Oil Prices Jun 19, 2026 Markets Rally as Iran Peace Sparks Chip-Led Surge; Oil, FX and Gilts Watchlists Shift Jun 19, 2026 U.S. Vice President Vance Cancels Switzerland Meeting with Iranian Delegation Jun 19, 2026 Pentagon Seeks $80 Billion Supplemental for Iran Conflict, Domestic Spending Jun 18, 2026 U.S. Trade Action Targets German Pharmaceutical Pricing Reforms Jun 18, 2026