With the public release of the "Mythos" large model that caused "cybersecurity panic" about to begin, Anthropic, the leader in large AI models, announced on Tuesday that it would open a preview of the cutting-edge model to about 150 new institutions to scan and repair major security vulnerabilities in network infrastructure in advance.
(Source: Company official website)
Anthropic previously disclosed,Since being open to preview to 50 initial partners in April, the "Myth" large model has helped these organizations find more than 10,000 high-risk or serious security flaws, and the speed of discovering vulnerabilities has increased "more than ten times.".
The company said most of the 150 new institutions expanded on Tuesday provide critical infrastructure services to their respective countries, such as power, water, medical, communications and hardware industries. In addition, many of Anthropic's key suppliers have also received early preview access.
In a statement, Anthropic said what these organizations have in common is that a successful attack on their code base could have catastrophic consequences.For most partners, a major attack could impact more than 100 million people.
It is reported that South Korea’s Samsung, SK Hynix, and SK Telecom have all obtained early preview qualifications. In addition, NATO and the EU cybersecurity agency Enisa are also included in this round of lists.
In addition to detecting major security vulnerabilities in network infrastructure, the "Myth" large model also has industry-leading programming capabilities. Anthropic said,Because advanced large models can uncover a large number of vulnerabilities, the bottleneck of network security has become verifying, disclosing and fixing the rapidly emerging high-risk vulnerabilities.
As the only non-U.S. entity known to have gained access among the 50 institutions in the first round, the British Artificial Intelligence Security Institute (AISI) evaluated the "Myth" large model and said that the model is capable of carrying out complex cyber attacks that "require days of work by human professionals." AISI also pointed out in a blog post last month that it is unclear whether the "mythical" large model is "an individual breakthrough that is out of touch with the established rate of progress" or "part of a faster development trend."
Anthropic expects,Within the next 6 to 12 months, many other AI companies will have models similar to “Myth”. In that case, cyberattacks are likely to be more frequent and less predictable.
It is worth noting that the public release of the "Myth" large model has been put on the schedule.
In last week’s announcement of the latest period model Claude Opus 4.8, Anthropic mentioned,The "Myth" level model is expected to be available to all users "in the coming weeks".
Trading data from the prediction market Polymarket shows that traders believe that the probability of the "Myth" large model being released this month is only 26%, but the probability of being released before the end of July reaches 67%.
In addition, there are market reports that,The performance of OpenAI's upcoming GPT-5.6 is expected to be close to the level of the "mythical" large model, and at the same time "the cost of use is 2-3 times lower."
Anthropic also announced on Monday that it would submit a confidential prospectus to the U.S. Securities and Exchange Commission. According to general experience, if there are no major environmental changes in the U.S. stock market, Anthropic’s IPO may be launched this fall. Obviously, whether the "mythical" big model is really as magical as the company promotes will be a potentially key factor in determining Anthropic's post-IPO trading sentiment.