Claude Mythos from Anthropic has shown just how powerful AI security tools can become.
The AI model can detect software vulnerabilities so effectively that Anthropic has not released it to the general public. Instead, access remains limited to security researchers and large organizations.
The reason is clear. A tool this strong could help defend systems, but it could also become dangerous if criminals use it.
AI helps uncover an Apple security flaw
According to the article, researchers from Calif used Claude Mythos Preview to investigate a serious security issue in Apple software.
The issue involved deep memory-level behavior inside the macOS kernel working with Apple’s powerful M5 chip.
In simple terms, the flaw could allow a regular user without high-level access to take full control of a device.
That makes the discovery extremely serious.
Claude Mythos guided the research
The most impressive part is how the AI helped the researchers.
Claude Mythos Preview reportedly guided the team during the investigation and helped develop testing code for the exploit.
Calif explained that the AI became especially useful because it could apply knowledge from one type of attack to similar problems.
That means once the model understands one class of vulnerability, it can move faster when facing related issues.
Apple responds to the report
The article notes that it is still unclear whether the exact flaw has already been fully fixed.
A macOS Tahoe 26.5 update reportedly mentioned a bug reported by Calif in cooperation with Claude and Anthropic Research.
However, Calif also said it recently met with Apple and will reveal more technical details after Apple finishes closing the attack path.
Apple also sent a statement to The Wall Street Journal saying user security remains its top priority.
AI security tools need strong limits
This case shows why advanced AI security models are both exciting and risky.
They can help researchers discover dangerous flaws faster than before.
However, if the same tools reach the wrong hands, they could also make cyberattacks easier.
THIS IS our take
Claude Mythos sounds like the kind of AI tool that makes cybersecurity teams cheer and panic at the same time. Finding flaws faster is great, but the industry now needs strict guardrails before AI-powered hacking becomes too easy.
SOURCE: Mashable
