Script Kiddies Turn Lethal

In the wake of Anthropic's Claude Mythos, a new era of AI-enhanced amateur hacking is poised to emerge, threatening to redefine the cybersecurity landscape.

The burgeoning capabilities of AI were underscored last August at DARPA’s Artificial Intelligence Cyber Challenge (AIxCC) in Las Vegas. Elite cybersecurity teams showcased their AI bug-finding systems, which meticulously scanned 54 million lines of software code intentionally seeded with flaws by DARPA. While the tools adeptly identified most of these artificial vulnerabilities, their prowess extended further, uncovering over a dozen genuine bugs that DARPA had not introduced.

Even prior to the industry-wide tremor caused by Anthropic's Claude Mythos this month – an advanced AI model seemingly capable of pinpointing vulnerabilities in any software it analyzes – automated systems were already demonstrating increasing proficiency in detecting coding flaws. A growing apprehension now exists that AI will not only find these weaknesses but also be weaponized to exploit them, effectively democratizing sophisticated hacking skills globally.

As one expert starkly put it, "Mythos or not, this is coming."

This is not an idle warning. For decades, "script kiddies" – individuals lacking deep technical expertise – have caused significant disruption by deploying pre-written scripts obtained from the internet or exploit toolkits. Despite their limited understanding of the underlying code, they successfully defaced websites and disseminated viruses.

However, the current situation represents a significant escalation. Individuals without extensive technical backgrounds can now leverage AI to dramatically amplify their capabilities in ways far beyond what simple scripts allowed. This shift is anticipated to have profoundly widespread repercussions.

Dan Guido, CEO and cofounder of cybersecurity firm Trail of Bits, a runner-up in the AIxCC challenge, articulated the impending shift: “There’s a tidal wave coming. You can see it. We can all see it.” He challenged the industry, asking, “Are you going to lay down and die, or are you going to do something about it?”

Anthropic, cognizant of the potential for misuse, is actively working to prevent criminals from exploiting its software, extending efforts beyond initiatives like Project Glasswing. A week after the Mythos announcement, the company released Claude Opus 4.7, integrating safeguards designed to block malicious cybersecurity requests. Security professionals seeking to utilize the model defensively can apply to Anthropic’s Cyber Verification Program.

While Mythos sent shockwaves, earlier indicators of AI’s cybersecurity prowess were evident. In June 2025, for instance, the autonomous offensive security platform XBOW surpassed human hackers to claim the top spot on HackerOne’s bug bounty leaderboard, signaling significant advancements in AI models' ability to discover vulnerabilities.

By the time the AIxCC commenced, Guido noted, "there were already 10 to 20 different bug-finding systems that could find orders of multitude more bugs than we could patch." He emphasized, "This is actually not a new problem."

A dire prediction looms: "2026 is the year when all security debt comes due… 2026 is the make-it-or-break-it year."

AI's exceptional pattern-matching abilities are making it progressively easier for individuals to identify variants of known bugs, as well as entirely new, undiscovered flaws. Concurrently, the process of writing exploits has become significantly simpler.

Tim Becker, a senior security researcher at Theori and an AIxCC finalist, highlighted this efficiency: “You can use AI tools and with very minimal human guidance, and in some cases no human guidance, find a zero day in widely used software.”

Industry-wide concern is palpable, fueled by the rapid pace of model improvements and a deeper understanding of their expanding capabilities.

The emergence of open-weight models—AI models with publicly available trained parameters—introduces further risk. Becker suggests that sophisticated threat actors are likely to deploy these models on their own infrastructure to prevent their exploits from being detected on platforms like Anthropic or OpenAI, which may monitor for abuse. The industry is bracing for potential releases from other model creators who may not exercise the same caution as Anthropic, potentially unleashing powerful new tools directly to the public.

Echoing his earlier warning, Guido reiterated, "Mythos or not, this is coming."

While Mythos represents a significant leap in exploit generation, existing models also possess considerable capabilities. Security researchers are already leveraging widely available models to proactively report vulnerabilities to vendors before they can be exploited in the wild. Conversely, this also creates a risk: malicious actors could utilize these same models for nefarious purposes, such as developing exploits for oppressive regimes or independently stealing sensitive data.

Industry experts foresee that advancements in AI's security capabilities will lead to a substantial increase in exploits. Malicious actors could direct AI to uncover bugs in obscure software components that previously would have been too labor-intensive for human exploitation.

As one expert observed, “The bar to diving into a new million-line codebase and finding a bug is so much lower than it used to be.”

Guido elaborated on this paradigm shift: “Now, because effort is cheap, you can do things that are lower down the food chain. You can write exploits for software that only one company has. You can write exploits for software that exists in only one configuration that one company has. And you can do it on the fly. So during the middle of an intrusion into some hospital and there’s a wall standing between you and what you want, you can just point an LLM at that wall and say, ‘Figure out a flaw here,’ and it can grind until it’s successful. And it’ll find some vulnerability, it can find some configuration, it’ll run an exploit, for a weakness that no one ever has before, and it’ll do it with almost no effort on the part of the user… the hacker… the script kiddie.”

This development, Guido argues, effectively "supercharges" script kiddies. They will be able to operate with agility, no longer constrained by the need to memorize weaknesses in obscure UNIX utilities, but rather relying on the AI tool's inherent pretraining. This enables them to iterate through exploit attempts at machine speed, a feat impossible for any human, let alone a traditional script kiddie.

While precisely quantifying the extent to which