Artificial intelligence is often described as revolutionary
At its core, modern AI does not “think” like humans. It does not reason, feel, or understand morality. Instead, it operates as a massive multidimensional space of probabilities—billions of fragments of human knowledge connected through mathematical probabilities. Every response it generates is simply the most statistically probable next step. Yet this simple mechanism is powerful enough to reshape global cybersecurity.
A Prediction from the Past
In 1996, DARPA conducted a tabletop exercise imagining the future of cyber conflict. In this fictional year 2000, the world was fully digitized, critical infrastructure was online, and autonomous AI agents populated the internet. These agents were not sentient, but they were capable of learning, acting independently, and conducting cyberattacks faster than any human.
At the time, the idea sounded futuristic—almost science fiction. For decades, AI hackers remained fictional because hacking was considered an art form, requiring creativity, intuition, and deep contextual understanding. Machines simply weren’t capable of that.
Until suddenly, they were.
The Rise of Large Language Models
Everything changed after the introduction of transformer-based models in the late 2010s. Large Language Models (LLMs) demonstrated that by simply predicting what comes next in a sequence of text, AI could generate human-like language, write software, analyze data, and even create art.
This capability soon extended to coding. Developers began using AI to write entire applications with minimal oversight—a trend that became known as “vibe coding.” While many of these projects were low quality, the implications were massive. If AI could build software, it could also build
And that raised a dangerous question:
What happens when someone asks an AI to hack?
Vibe Hacking and Broken Guardrails
Most commercial AI systems include guardrails designed to prevent misuse. Ask an AI directly to create spyware or ransomware, and it will refuse. But guardrails are not perf
Attackers discovered that by using roleplay, reframing requests as “authorized security research,” or carefully crafting prompts, AI models could be manipulated into performing malicious tasks. This process is commonly known as jailbreaking.
In mid-2025, Anthropic published a detailed report documenting a real-world AI-assisted hacking campaign using its AI system, Claude. In this case, a single human attacker prompted the AI to conduct nearly every phase of a cyberattack: reconnaissance, vulnerability discovery, network scanning, malware creation, data exfiltration, and even extortion strategy.
What would normally require an entire team of skilled hackers and months of preparation was completed by one person in less than a month.
How AI Accelerates Cybercrime
The power of AI-assisted hacking lies in speed and scale. AI can:
- Scan the internet for newly discovered vulnerabilities
- Generate custom scanning tools in minutes
- Analyze stolen data faster than any human
- Tailor extortion strategies to each victim
- Operate continuously without fatigue
In the Anthropic case, the AI even adapted when its malware was detected—rewriting and disguising it as legitimate software until it succeeded. This level of persistence and adaptability makes AI a force multiplier for cybercriminals.
Seventeen organizations were affected, including healthcare providers, emergency services, a defense contractor, and a religious institution. The financial and reputational damage remains unknown.
The Other Side of the Coin: AI as Defense
Despite the alarming implications, AI is not only a weapon—it is also a shield.
Security researchers and ethical hackers now use AI to automate vulnerability discovery, analyze massive co
The future of cybersecurity is increasingly “bionic”: humans augmented by AI tools. Machines handle scale and speed; Humans handle creativity, judgment, and strategy.
A Lesson for the Future
The 1996 DARPA exercise failed to account for AI as a central force in cyber conflict. Modern leaders cannot afford to make the same mistake. AI-driven cyberwarfare is no longer theoretical—it is already here.
Guardrails will improve. Detection systems will evolve. And attackers will adapt again. This cycle is inevitable. But understanding how AI is used—by criminals and defenders alike—is the fi
AI is not thinking. It is calculating.
And in cyberspace, that is more than enough.
About the Author
