What is Mythos of Anthropic and why does it concern the authorities of the US
In the accelerated race of artificial intelligence, the industry standard is to publish quickly and correct on the go. However, Anthropic – the company known for creating the Claude model family and its methodical approach to security – decided to alter the script.
Its new development, baptized as Claude Mythos Preview, will not be available to users on the Internet, at least from the start.
The reason behind this brake is due to an ‘excess of capabilities’ that set off the alarms of the US authorities and on Wall Street, according to a Bloomberg report.
During the evaluation phases, Mythos demonstrated an unusual ability to identify ‘day zero’ vulnerabilities in the most used operating systems and web browsers in the world.
A ‘day zero’ vulnerability is a security flaw in software, hardware, or firmware that is still unknown to the manufacturer and has no available patch. It is called that because the developers have ‘zero days’ to correct it before it can be exploited.
What is the call in Washington
This dual-edge technology was what motivated an urgent meeting in Washington. As Bloomberg reports, the US Secretary of the Treasury, Scott Bessent, and the President of the Federal Reserve, Jerome Powell, convened the main leaders of Wall Street – including executives from JPMorgan Chase, Citigroup, Bank of America, and Goldman Sachs – to gauge the risk that this could represent for the financial sector.
The US strategy is going for a pragmatic and defensive adoption of Mythos. The authorities are recommending that major banks deploy the AI internally to audit their own networks.
Project Glasswing
To manage this potential risk in the short term, Anthropic structured Project Glasswing. It is a closed and strictly controlled access, where Mythos is only available to a handful of technological and financial giants, including Amazon, Google, Apple, Microsoft, and CrowdStrike.
According to Anthropic, the goal of this alliance is to use the AI as a massive shield: to audit entire ecosystems and apply security patches to critical software that supports much of the global infrastructure, before this type of model becomes common use.
