The security balance in the global artificial intelligence field is undergoing a delicate shift. On Tuesday, the renowned AI lab Anthropic officially announced that it would expand the access to its strictly controlled top model, Mythos, adding approximately 150 institutions with access. This move has increased the total number of institutions authorized to call the model to around 200.

In this new list, Anthropic deliberately included vertically underrepresented industries. According to official disclosure, these unnamed institutions are spread across key sectors such as power, water, communications, healthcare, and hardware manufacturing. Many of the newly added partners come from more than 15 countries, and they are mostly suppliers of underlying code libraries that are deeply relied upon by governments and major national infrastructure.

Distributing such a powerful model to critical institutions is primarily aimed at defense. Anthropic stated that this move is intended to enable companies managing critical infrastructure to use Mythos to identify and reinforce their cybersecurity defenses in advance. Before this, the first group of about 50 partners (including the US government, investment banks like JPMorgan and Goldman Sachs, and tech giants like Amazon, Apple, NVIDIA, Broadcom, and CrowdStrike) had already used the model to uncover over 10,000 "high-severity or critical" system security vulnerabilities.

However, this action has also raised considerable concerns in the industry, as Mythos itself is a highly sharp double-edged sword. According to reports, the model has extremely advanced cybersecurity and hacking capabilities, and if control is lost, it could pose a devastating threat to national security or critical infrastructure. Anthropic assessed that the consequences for most partners on the list, if they were successfully attacked, would be catastrophic, potentially affecting over 100 million people worldwide.

Facing the criticism of "raising tigers to guard the house," Anthropic believes it is an unavoidable technological confrontation. They predict that many other AI companies will develop models of comparable scale within the next 6 to 12 months, and it is highly likely that they will be released directly to the public without anti-abuse safeguards. Therefore, they hope that through this initiative called the "Glass Wing Project," critical institutions can establish an industry operational standard to deal with this reality before the danger truly arrives.

Notably, the day before opening model access, Anthropic secretly submitted public offering documents to the US Securities and Exchange Commission, preparing to join SpaceX and OpenAI in launching the largest initial public offering (IPO) wave of the year. Last week, the company just completed a $6.5 billion H-round funding, pushing its post-money valuation to $96.5 billion. Its recent annual recurring revenue has already exceeded $4.7 billion. As one foot steps into the capital market, how to balance the destructive power and defensive capability of cutting-edge technology has become a long-term challenge for this new giant.