“We must open a new era where individuals can use the powerful new capabilities of artificial intelligence (AI) to reach their full potential, pursue their interests, and create a better life and world than before.”
Mark Zuckerberg, Meta chief executive officer, stated this in an essay titled “The Future is for Everyone” released on Aug. 10 (local time), saying, “Artificial Superintelligence (ASI), which will be one of the most important technologies in history, must be widely deployed so that everyone can control it.” This stands in opposition to the “closed AI” approach maintained by companies like OpenAI and Anthropic.
Meta previously developed and advanced “Llama” as an open-source-centered AI model, but shifted its strategy to closed AI last year in consideration of market conditions centered on closed AI. However, by recently releasing “Muse Glimmer,” an open-source AI model usable on personal devices, it declared its return to an open ecosystem.
Zuckerberg proposed a philosophy that ASI should take personal empowerment as the source of prosperity, invention as the primary purpose of superintelligence, and the balance of power as the foundation of safety.
Zuckerberg emphasized that a world where everyone has an ASI lawyer will be fairer and more efficient than a world where only one person has an ASI lawyer. He warned that if a single specific company holds ASI, prosperity will be limited.
He also criticized, “I cannot understand why people who believe AI will eliminate most jobs and make humanity lose much of its reason for existence are in a hurry to build such a future.” This is interpreted as a remark targeting companies like Anthropic and OpenAI, which warn of AI risks while being the most active in closed AI development.
Zuckerberg takes the position that reviews of risks such as government cybersecurity or misuse in biochemical weapons development should also be actively conducted. As an alternative, instead of governments reviewing models after they are ready for launch, he suggested providing governments with intermediate training checkpoint access for new frontier models and technical personnel.
The intent is that governments can secure capabilities without restricting or delaying individuals' access to ASI and without causing power imbalances. He pointed out that if systems such as prior authorization pursued by the current Trump administration cause a one-month delay in launching frontier AI models in the U.S., the technological development pace of overseas models from places like China could surpass that of the U.S.
He also stated that distillation-based training, which has been defined as an act of infringing on intellectual property (IP) by referencing answers from other AI models, should be approached openly. U.S. companies such as Anthropic and the U.S. government maintain a firm stance, including bans, against distillation-trained models from Chinese AI companies like DeepSeek.
Meanwhile, Muse Glimmer, unveiled by Meta on the same day, is a small model specialized for on-device AI with 30 billion parameters. It can run agent tasks around the clock with just a single graphics processing unit (GPU) inside a standard PC. According to The Wall Street Journal, Meta also plans to introduce an open-weights version of its top-tier model, “Muse Spark 1.2,” within a few weeks.
To dispel AI safety controversies, Zuckerberg established a new governance structure granting an internal independent board the authority to deliberate on AI model safety standards and control whether to launch them. Meta also decided to create a $1 billion (approximately KRW 1.4 trillion) “The Future is for Everyone” fund to build AI infrastructure and support local communities in the U.S.
