Joey Melo, a prominent figure in cybersecurity, approaches hacking with a unique perspective focused on controlling environments without altering fundamental rules. This philosophy traces back to his childhood interest in the video game Counter-Strike, where he enjoyed modifying in-game elements, such as character speeds and uniform colors, rather than following traditional gameplay.
From Pentesting to AI Red Teaming
Currently, Melo serves as a Principal Security Researcher at CrowdStrike, having transitioned from roles at Bulletproof and Packetlabs to becoming a red team specialist at Pangea. His shift from pentesting, which is usually more focused, to the broader scope of AI red teaming reflects his growing interest in artificial intelligence. His self-driven exploration into AI began as a side project while he was still a pentester.
In March 2025, Melo participated in an AI hacking competition organized by Pangea while working at Packetlabs. His obsessive dedication led him to win the competition, which further fueled his move to Pangea as a red team specialist. Melo’s background in pentesting provided a strong foundation, but his natural curiosity and desire to manipulate environments for learning and fun have been equally influential in his career.
Understanding AI Jailbreaking
Jailbreaking AI involves bypassing its guardrails to manipulate outputs without modifying the source code. Melo explains that this process requires understanding the AI’s intended functions and constraints. By crafting specific prompts, he can explore the AI’s capabilities and limitations. This helps in identifying how contextual manipulation can lead to successful jailbreaks.
For instance, by altering the context in which an AI operates, such as simulating a future year where certain activities are legal, Melo can sometimes convince the AI to bypass its guardrails. This requires creativity and an understanding of how AI retains conversational context, allowing hackers to subtly shift the AI’s operating parameters.
The Role of Data Poisoning
Beyond jailbreaking, Melo also investigates data poisoning, where input data is manipulated to cause incorrect outputs. This can lead to significant consequences, especially in critical systems like medical diagnostics or autonomous vehicles. Melo uses adversarial techniques to test AI models’ susceptibility to such attacks, often by creating misleading data that the AI might ingest during its training process.
Despite the challenges posed by evolving AI technologies, Melo emphasizes the importance of ethical hacking in strengthening AI security. His responsible disclosure of vulnerabilities helps developers improve AI guardrails, contributing to a more robust security landscape.
Melo’s commitment to ethical practices in cybersecurity remains steadfast, reinforcing the idea that the ability to cause harm should be tempered by a conscious choice to act responsibly. His work continues to influence the field, ensuring that AI systems become progressively more secure against emerging threats.
