AI Cyber Threat Emerges Models Expose Software Flaws

• Existing AI models can replicate Mythos' vulnerability findings.
• AI-enabled hacking poses significant threats to corporations and governments.
• Defensive measures lag behind offensive AI capabilities in cybersecurity.
• Limited access to advanced AI models hinders collaborative cybersecurity innovation.

A Disturbance in the Force A New Cyber Threat

I sense a great disturbance in the Force. Global banks, tech giants, and even those bumbling bureaucrats in government are in disarray over Mythos, the Anthropic model that claims to have discovered thousands of software vulnerabilities. It seems the ability to find weaknesses is now considered a weapon. Perhaps I should have used this to find that exhaust port sooner. But here's the twist the power they fear is already here. Cybersecurity experts claim that existing models, including those from Anthropic and OpenAI, can achieve similar results. "What we are seeing across the industry now is that people are able to reproduce the vulnerabilities found with Mythos through clever orchestration of public models to get very, very similar results," says Ben Harris, CEO of watchTowr. Clever orchestration, indeed. It's like those Rebel pilots coordinating their attack runs on the Death Star, only this time, the target is the entire digital galaxy.

The Emperor's New AI Oversight

Fear leads to anger, anger leads to hate, and hate leads to government oversight. The release of Mythos, albeit limited to a select few American companies like Apple, Amazon, and JPMorgan Chase, has prompted the Trump administration to consider new regulations. How predictable. Even in this digital age, the desire to control and dominate remains. Weeks after Mythos' arrival, OpenAI CEO Sam Altman announced GPT-5.5-Cyber, a model tailored for cybersecurity. Another pawn in this game of galactic chess. The controlled rollout of Mythos, dubbed Project Glasswing, aims to give the corporate world time to strengthen its cyber defenses. A futile effort, perhaps? The dark side is always stronger, especially when it comes to exploiting vulnerabilities. Especially given the discussion taking place around AI Data Centers and the challenges they are creating, it is important to also mention that similar power issues, mushrooming of power usage and vulnerabilities are also related to this topic as highlighted in AI Data Centers Spark Power Price Panic Mushrooms and Megawatts.

Zero-Days The Ultimate Weapon

The ability to detect zero-days at scale is now within reach, and Klaudia Kloc, CEO of Vidoc, confirms that existing models have possessed this capability for months, if not a year. This is indeed "scary enough." A zero-day vulnerability is a previously unknown flaw that hasn't been patched, providing attackers with a window of opportunity. It's like finding a weakness in the Death Star's defenses before the Empire even realizes it's there. Vidoc leveraged a technique called "orchestration" to test if they could replicate Mythos's findings. By splitting code into smaller pieces and coordinating between various tools and models, they successfully detected the same vulnerabilities using older models from OpenAI and Anthropic.

A Thousand Detectives Are Stronger Than One

The Empire relies on overwhelming force, but sometimes, a clever strategy is more effective. Aisle found that many of Mythos's results could be reproduced using cheaper models working in parallel. Scale and coordination are more important than having the latest, most powerful model. "A thousand adequate detectives searching everywhere will find more bugs than one brilliant detective who has to guess where to look," wrote Aisle founder Stanislav Fort. A lesson the Empire could learn from. Throwing more resources at a problem doesn't always guarantee success. Sometimes, a more strategic approach is required.

The Hysteria Is Strong With This One

Corporations and government regulators are panicking over the threat of AI-enabled hacking, according to Harris. He described recent conversations with banks, insurers, and regulators as "hysteria." Even before generative AI, companies struggled to patch vulnerabilities faster than skilled hackers could exploit them. The barriers to cyber havoc have been lowered, meaning more attacks and threats to previously uninteresting systems. It seems fear of the unknown is a powerful motivator, even in the digital realm. This hysteria reminds me of the fear I sensed in the Rebel base before the attack on the Death Star.

The Offense Has the High Ground

While Anthropic, OpenAI, and others are working on cyber defense capabilities, the initial advantage lies with the offense. JPMorgan's Jamie Dimon suggested that AI tools are making companies more vulnerable before they can help defend them. "You have a significant increase in the volume of vulnerabilities discovered, but they don't seem to have deployed a tool that helps you fix them," said Justin Herring, partner at Mayer Brown. A vulnerability management is described as Sisyphean task of cybersecurity. The limited group that was part of the initial Mythos release got a head start on patching vulnerabilities, but this prevents the wider cyber community from being part of the solution. Many cybersecurity startups are working on solutions to fix the world before this becomes accessible to the world. A chicken-and-egg situation, indeed. It seems inevitable that some eggs will be broken along the way.