Antitrust worries for Anthropic
- Commentators flagged Anthropic’s 'Project Glasswing' as potentially anticompetitive if access to its model is exclusive for critical infrastructure security. - ProMarket argued that exclusivity in such a strategically important layer could entrench market power and raise antitrust risks. - The debate shows antitrust scrutiny is extending into who controls model access, data, and security capabilities within the AI stack (promarket.org)
Anthropic’s Project Glasswing is drawing antitrust scrutiny because Anthropic is giving early access to a powerful cyber model to a closed group of major companies. (anthropic.com) Anthropic announced Project Glasswing on April 7, 2026, saying launch partners include Amazon Web Services, Apple, Broadcom, Cisco, CrowdStrike, Google, JPMorganChase, the Linux Foundation, Microsoft, NVIDIA and Palo Alto Networks. Anthropic said those partners are using Claude Mythos Preview for defensive security work. (anthropic.com) Anthropic also said it extended access to more than 40 additional organizations that build or maintain critical software infrastructure, and committed up to $100 million in usage credits plus $4 million in donations to open-source security groups. (anthropic.com) The model at the center of the program is not public. Anthropic said Claude Mythos Preview had already found thousands of high-severity vulnerabilities, including some in every major operating system and web browser, before the company decided to limit access. (nextgov.com) Anthropic and other security outlets described the system as a model that can read complex code, spot hidden flaws and sometimes suggest or carry out fixes, which is why the company framed the project as a defensive effort for critical software. Infosecurity Magazine reported Anthropic publicly identified examples including a 27-year-old OpenBSD flaw and a 16-year-old FFmpeg flaw. (infosecurity-magazine.com) The antitrust argument came into focus on April 22, when ProMarket published a commentary by Madhavi Singh arguing that a consortium with exclusive access to a strategically important model could raise issues under Section 1 of the Sherman Antitrust Act. Singh wrote that information-sharing rules and exclusion of outsiders could look like concerted action if the group controls critical security knowledge. (promarket.org) That critique is aimed less at consumer chatbots than at control over a layer of the artificial intelligence stack: who gets the model, who gets the vulnerability data, and who gets to harden essential software first. Anthropic, for its part, said it will “share what we learn so the whole industry can benefit,” even while keeping Mythos Preview restricted. (anthropic.com, promarket.org) Supporters of the restricted rollout say the risk is immediate. Nextgov/FCW reported Anthropic briefed senior United States officials, including people connected to the Cybersecurity and Infrastructure Security Agency and the National Institute of Standards and Technology, before any external release because the same model could aid offensive cyber operations if it spread widely. (nextgov.com) Senator Mark Warner, a Virginia Democrat, praised the April 7 launch and said cyber threat actors are already using artificial intelligence tools to improve their capabilities. His statement backed faster patching of critical systems, not broader public release of the model. (warner.senate.gov) The dispute leaves Anthropic defending two positions at once: that Glasswing is narrow enough to keep a dangerous model out of the wrong hands, and broad enough that it does not lock a small circle of firms into a privileged security role. (anthropic.com, promarket.org)
