A strange game is being played between the cybersecurity laboratories and the White House offices: that of an artificial intelligence that, at the same time, promises to protect the Internet… and demonstrates how easy it would be to break it.
The system is called Mythos, and it is the most advanced model yet developed by Anthropic (the same company that publicly disagreed with Trump’s policy of using autonomous and lethal systems). On paper, Its function is almost noble: finding vulnerabilities in software before attackers do. In practice, its capabilities have proven uncomfortably ambiguous.
During its first tests, Mythos was able to detect, and in some cases exploit, security flaws in virtually all relevant systems: operating systems, browsers, critical infrastructures. Some of these failures had been hidden for decades. That is the turning point. It is not just about AI finding errors, something that other tools already did, but that it can go through the entire path: detect the crack, understand it and turn it into a functional entry door. From there, the narrative branches.
On the one hand, there is the promise. Technology companies and selected organizations have begun to use Mythos within a closed program (Project Glasswing) with a clear idea: let AI scan the system before attackers, discover massive vulnerabilities and fix them. In this scenario, Mythos functions as a kind of relentless auditor, capable of reviewing in weeks what would take a human team years.
But at the same time, the awkward mirror appears: any tool that can find bugs can also exploit them. And do it, furthermore, at a speed that has no human equivalent. Some experts already warn that these types of models do not invent new forms of attack, but they do accelerate existing ones to a critical point. And that is one of the advantages/problems that the most advanced AIs face: They are able to navigate through millions of data, until they find the needle in the haystack.…And do it instantly. What would take us lives takes seconds for them: it is the instantanthropyorder chaos in an instant.
While some agencies linked to security have shown reluctance, the White House has opted for another approach: incorporate Mythos, with restrictions, into the federal ecosystem. The logic is almost pragmatic: if this technology is going to exist, it is better to understand it from within than to confront it from the outside.
Gregory Barbaccia, federal CIO of the White House Office of Management and Budget (OMB), informed Cabinet department officials yesterday by email that the OMB is implementing safeguards that will allow its agencies to begin using the Mythos AI toolthe use of which is kept strictly confidential.
The email does not confirm that the various agencies will have access to Mythos, nor does it provide a timeline for its availability or how they will use it. Indicates to the main technology and cybersecurity managers that they will receive more information “in the coming weeks”. The result is a paradoxical scenario. On the one hand, governments and companies compete to use Mythos to strengthen their security. On the other hand, this same deployment reveals the extent to which current security is fragile. Each vulnerability discovered is not only a problem solved, but also a demonstration of what could have gone unnoticed.
Ultimately, the question Mythos poses is not technological, but strategic. What happens when defense and offense use exactly the same tools? And what happens when those tools evolve faster than the ability to regulate them?