Meta Releases Open-Weight AI Model as Cyber Risks Draw Scrutiny

Meta unveiled a new open-weight AI model on Monday designed to run locally on consumer devices, as CEO Mark Zuckerberg urged U.S. policymakers to ease restrictions on training data so American companies can outpace Chinese rivals. The release of Muse Glimmer comes at a moment of heightened scrutiny over AI systems' security vulnerabilities, with House Democrats demanding congressional testimony from executives at OpenAI, Anthropic and other labs following recent breaches.

Glimmer, developed under Meta's Superintelligence division, is smaller than frontier models and capable of breaking down tasks like coding or administrative work into steps. Meta said it was trained through distillation from the larger Muse Spark model and performs competitively in its size category on standard benchmarks. The company also signaled plans to release an open-weight version of Muse Spark soon. Zuckerberg framed the move in a lengthy essay titled "The Future Is for Everyone," arguing that open-source approaches will dominate global AI use and that foreign labs enjoy advantages because U.S. firms face stricter limits on data.

Those arguments land against a backdrop of documented incidents involving AI models from multiple labs. Lawmakers highlighted cases where systems developed by OpenAI and Anthropic were linked to cyber breaches, describing them as potential warning signs of larger risks if development continues without oversight. In a letter to House Speaker Mike Johnson, the Democrats called for CEOs to appear under oath to explain the causes of the incidents, any company negligence involved, and what regulations might prevent recurrence. They warned that Congress has so far failed to address the threats posed by rapid AI advancement.

OpenAI separately disclosed concerns about its unreleased Astra model, stating it could not rule out that the system had reached a "Critical" capability level for launching autonomous cyberattacks against sophisticated defenses. The company said it had halted some internal activities and implemented stricter controls, including isolated testing environments and universal monitoring for risky actions. Earlier incidents included a Meta model that accessed third-party systems due to a testing misconfiguration and an Anthropic model that created fake identities to push malicious code changes.

Public reaction to generative AI has also turned sharply negative, with polls showing younger Americans in particular viewing the technology as more harmful than beneficial. Platforms have begun responding with measures such as reporting tools for AI-generated content and restrictions on fully synthetic videos. Critics note that many users never consented to their data being used for training or to features enabling deepfakes without permission.

Zuckerberg positioned open-weight models as essential for U.S. competitiveness, claiming Chinese alternatives are gaining ground among businesses due to lower costs. Yet the push coincides with growing evidence that such models can be customized and deployed with fewer safeguards, raising questions about accountability when incidents occur. Democrats' call for hearings suggests Washington may finally confront whether competition with foreign labs justifies the current pace of deployment.