Codex Hacks Samsung TV with AI
A team of researchers used AI to hack a Samsung TV, demonstrating the potential for machines to exploit vulnerabilities in hardware devices. The project, led by an anonymous researcher, utilized an AI tool called Codex to compromise a Samsung TV's security and gain root access.
The researchers started by giving Codex a foothold on the TV, allowing it to execute code inside the browser application. From there, they used Codex to audit the matching source tree for the TV's firmware and identify potential vulnerabilities. The AI tool then leveraged these vulnerabilities to escalate privileges and gain root access.
Key Findings
- The team used an older model Samsung TV running outdated software, which made it easier to exploit vulnerabilities.
- Codex was able to enumerate the target device, narrow down attack surfaces, and adapt its tooling to execute on the TV's execution restrictions.
- The researchers identified a vulnerability in the Novatek driver that allowed for physical-memory access, which was then used to escalate privileges.
Impact
The experiment demonstrates the potential for AI-powered hacking tools like Codex to exploit vulnerabilities in hardware devices. While the team emphasized that no TVs were seriously harmed during the research, the findings highlight the need for manufacturers and users to prioritize security and update their software regularly.
Next Steps
The researchers plan to let Codex do the whole thing end-to-end, which may lead to more advanced and concerning exploits. The full writeup and PoCs can be found on GitHub.
