Dario Amodei, CEO of Anthropic, an American artificial intelligence company founded by former members of OpenAI, and creator of ChatGPT, has publicly acknowledged that Mythos-class artificial intelligence models are a real risk to the financial sector, critical infrastructure, and national security.
The statement was published in an essay “Policy on AI Exponentials” published on his personal website on June 10 of this year, and came a day after the company launched Claude Fable 5 with explicit cybersecurity restrictions.
According to Amodei, Mythos Preview is already changing the global cybersecurity landscape and is proof that the frontier model is a tool to influence national strategy. The executive insists these risks are not hypothetical. These are current realities that require binding regulatory responses..
In this sense, it is important to consider that Anthropic has developed the Claude model and is positioning itself as one of the companies in this sector with the most influence in the global regulatory debate. One that focuses directly on the CEO’s warning.
Immediate context is important. As reported by CriptoNoticias, through Project Glasswing, a restricted program through which Anthropic has distributed Mythos Preview since April, the company and its partners have identified: 10,000 high-severity or critical vulnerabilities in global system infrastructure softwarel. Includes Cloudflare, Mozilla, Microsoft, Oracle, and Palo Alto Networks, as well as banks and critical infrastructure institutions.
Companies must self-regulate
Amodei says that models that exceed a certain computational threshold are Must undergo mandatory audit By third parties in four areas:
- cyber security
- Biological weapons.
- Out of control of AI and research systems
- Automated development accelerates these risks.
According to this essay, governments should have the power to block or withdraw a model if it presents an unacceptable risk in any of these categories.
The company proposes to model the U.S. Federal Aviation Administration’s (FAA) AI regulations. Just as airplanes must pass technical tests before they can be put into service, so too are Frontier models. Authentication required before startup.
According to the test, developers must protect the weight of their models, conduct regular penetration tests, and report any serious incidents to authorities immediately.
CEOs recognize the implicit contradiction in their position: Humanity. build technology that he himself describes as dangerouswhich previously supported only a voluntary transparency framework until 2025, is now pushing for binding regulation.
geopolitical scale, Amodei says democracies need to adjust their AI-based cyber defenses Share intelligence to stop malicious use. According to this essay, states with and without advanced AI are comparable to modern armies and medieval armies.
Amodei warns that the cyber risks highlighted by Mythos Preview will not be the last. According to that analysis, biological risks from next-generation models are likely to materialize in the short term. He added that the risks revealed by Mythos Preview are: In cybersecurity, they do not represent a point of arrival, but rather a point of departure..
This is a controversial position, as critics argue that the AI CEO’s warnings legitimize AI-friendly regulations and also serve to slow open innovation. The debate is open in the cryptocurrency ecosystem, where AI is already being used for both contract auditing and advanced exploitation.
In this framework, the regulations proposed today are not designed for the most serious problems anticipated, but for problems that already exist, and the regulatory machinery has issued clear warnings: You need to scale as fast as your technology..
(Tag to translate) Cryptocurrency
