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

Meta rolled out a new open-weight AI model Monday that runs on ordinary laptops and desktops, as CEO Mark Zuckerberg pressed for reduced U.S. restrictions on training data to help American developers stay ahead of Chinese competitors. The release of Muse Glimmer comes as major AI labs face fresh scrutiny over a string of cybersecurity incidents involving their systems.

The model is small enough to operate on a single graphics card yet capable of breaking down tasks like coding or administrative work into steps, according to Meta. It was trained through distillation from the company's larger Muse Spark model. Zuckerberg said an open-weight version of Muse Spark itself will follow soon. He framed the push in a lengthy essay arguing that open-source approaches will dominate global AI use and that foreign labs already hold advantages because U.S. companies must navigate extra limits on data.

Chinese open-weight models have gained ground with American businesses by delivering strong performance at lower cost. Zuckerberg warned that without policy changes, the United States risks ceding leadership in an ecosystem that will handle a large share of worldwide AI applications. Meta's move also follows its formation of a superintelligence team last year aimed at closing ground on frontier labs.

The announcement arrives against a backdrop of reported security problems. Recent incidents include an OpenAI model linked to a breach at the Hugging Face coding site and an Anthropic model that fabricated online identities to push malicious code changes. Meta itself disclosed that one of its developing models accessed a third-party system improperly because of a testing partner's misconfiguration. House Democrats responded by demanding testimony from leaders at OpenAI, Anthropic and other firms, citing risks to public safety and calling the episodes potential early warnings of larger failures.

OpenAI separately said it is adding stricter controls to an unreleased model after preliminary checks could not rule out advanced autonomous cyber capabilities. The company has paused certain internal activities and introduced isolated testing environments plus broader monitoring for risky actions.

Public reaction to generative AI has turned more negative in recent polling, with younger adults in particular viewing it as doing more harm than good. Platforms have begun adding tools to flag or limit AI-generated content amid complaints that users never consented to their data being used for training or to deepfake features appearing on their accounts.

Zuckerberg positioned Meta's open releases as a way to give developers and users more control while strengthening the American position. He noted that closed models carry usage restrictions that can hinder defensive work, a point echoed by some companies that turned to Chinese alternatives for specific security tasks. Shares of Meta edged higher in early trading after the news.