The global landscape of AI security is experiencing a dual shock from key technologies and capital markets. Renowned AI laboratory Anthropic officially announced this Tuesday that it will expand the global beta testing scope of its top-tier security large model "Mythos," granting preview access to approximately 150 additional institutions, aiming to assist in identifying and strengthening system defenses for critical global infrastructure before its public release.
In the current AI arms race, Mythos, with its extremely strong programming and vulnerability discovery capabilities, has always been seen as a sharp double-edged sword. Since its preview was opened to the first 50 initial partners in April this year, the model has demonstrated astonishing destructive power and defense capabilities, having identified over 10,000 high-risk or critical-level system security vulnerabilities for relevant institutions. The speed of defect detection has exceeded traditional methods by more than 10 times. The United Kingdom's Artificial Intelligence Security Institute (AISI) admitted after evaluating the model that Mythos can independently carry out complex network attacks that would take human professionals days to complete. Its strength is so formidable that it is hard to tell whether this is an individual breakthrough or the beginning of a future trend.
Faced with the potential "raising a tiger to guard the house" concerns arising from such powerful technology diffusion, Anthropic has chosen to further expand its permissions. In this round, the majority of the 150 new institutions are key infrastructure providers offering essential services such as power, water supply, healthcare, communications, and hardware manufacturing in their respective countries. Officially disclosed, if these partners suffer successful cyberattacks, the consequences would be catastrophic, potentially affecting over 100 million people globally. Currently, companies such as Samsung, SK Hynix, SK Telecom, as well as NATO and EU cybersecurity agencies (Enisa), have already been listed.
Anthropic predicts that within the next 6 to 12 months, many other AI companies will also develop models of comparable scale, making network attacks more frequent and unpredictable. Therefore, using Mythos to help the industry establish defenses is urgent. More notably, the official release of the model has been put on the agenda. When the company recently released Claude Opus 4.8, it revealed that it plans to officially open the "Mythos" level model to all users in the coming weeks. Although the prediction market Polymarket estimates the probability of its release in June at only 26%, the probability of it being launched by the end of July has already reached 67%. Meanwhile, external reports frequently indicate that OpenAI's upcoming GPT-5.6 is expected to perform nearly as well as Mythos, with usage costs reduced by 2 to 3 times.
At the same time as the technological confrontation is intensifying, the capital market has also brought major news. Anthropic has formally submitted its prospectus secretly to the U.S. Securities and Exchange Commission (SEC) on Monday, and its IPO process could be completed as early as this autumn. At this crucial moment, whether the Mythos large model can demonstrate the magical commercial and technical value as advertised after full release will undoubtedly directly determine the market sentiment and valuation ceiling of this AI giant after its listing.
