Artificial intelligence is no longer a side tool in cybersecurity. It is becoming part of the core contest between defenders trying to find vulnerabilities first and attackers trying to exploit them faster. The New York Times reported this week that new AI systems are pushing the security industry toward a world in which the defense against AI-powered attacks is increasingly more AI.

That shift is already visible in the market. The Verge reported that Anthropic's new Claude Mythos Preview model, deployed through Project Glasswing, has been used to identify large numbers of severe vulnerabilities across major operating systems and browsers. At the same time, the model is being kept out of general release because of concern over how easily such capability could be repurposed.

The concern is not limited to one company. Axios reported that Anthropic's latest model displayed enough offensive cyber ability during testing to raise alarms about what could happen if similar systems spread without tight controls. That warning reinforces a broader industry fear that the same leap in reasoning and coding performance that helps security teams can also compress the time it takes hostile actors to develop exploits.

The larger backdrop is that business and government leaders already expect AI to be the defining force in cyber risk this year. The World Economic Forum said in its 2026 cybersecurity outlook that AI is reshaping the field across three fronts at once: it expands the attack surface, strengthens defensive operations and improves the scale and precision of attacks. The report says institutions are now confronting a cyber environment in which AI capability is rising on both the defensive and offensive sides.