GEEKCON 2026 Returns to Shanghai: Hackers Demonstrate AI-Assisted Exploits Across Apps, Robots and Vehicles

GEEKCON 2026 returned to the CMG Media Cinema in Shanghai’s West Bund art district after last year’s open-air airfield venue. The move from an uncovered tarmac to a conventional screening hall signaled a deliberate shift: after exposing security issues to sunlight in 2025, the conference now aims to project future risks onto a screen before they materialize.

Organizers opened the event with footage from The Matrix, acknowledging that the film’s depiction of AI-controlled systems and human resistance now functions more as a documentary preview than science fiction. GEEKCON chairman Wang Qi (Big Bullfrog) described his earlier fear that rapid AI progress had rendered human hackers obsolete, nearly leading to cancellation of the event under the banner of a “final human hacker competition.” He ultimately reframed the question mark at the end of that theme, emphasizing that vulnerabilities exist independently of researchers but are eliminated only when discovered.

The conference retained its core DAF vulnerability exploitation challenge and 30+5 deep-dive sessions while adding two new AI tracks: the AI Hacker Extreme Challenge and the AI Hacker Debate. In the Extreme Challenge, six teams received undisclosed targets only on-site and had one afternoon to compromise them with AI assistance. Targets included a national-scale mobile application, office software, an open-source large-model inference framework, panoramic cameras, quadruped robots, and consumer companion robots. Within forty minutes, teams had breached AI infrastructure and office software; wireless attacks later allowed remote hijacking of robots and cameras. In total, eleven of the six target classes were compromised.

The debate polled attendees on whether AI would eliminate human hackers; roughly 40 percent foresaw the profession’s decline while 60 percent believed humans would continue exploring unknowns. Results from the two tracks were telling: the human-only DAF challenge awarded no first or second prizes, whereas every team in the AI track received recognition.

Live demonstrations underscored real-world impact. One winning entry (9.3 points) showed how an unknown flaw in a mobile payment app allowed an attacker to hijack transfers via a single link, redirecting funds and altering amounts while the victim’s device displayed a successful transaction. Other PoCs included remote takeover of drones bypassing geofencing, theft of vehicle credentials enabling drive-away attacks, compromise of powered-off delivery robots that could then be driven into obstacles, and silent installation of malicious apps on Android devices via QR codes. Researchers also detailed firmware and AI Agent permission flaws in Unitree GO2 and GO2-W robots that could escalate from a prompt to motor torque commands, plus new exploitation paths in fastjson 1.2.83 uncovered with AI tooling.

Organizers stressed that the real question is not whether human defenders remain necessary, but why products continue to ship with preventable flaws. By surfacing these issues on stage, GEEKCON seeks to raise the security baseline of intelligent devices and ensure that the white-hat researchers who find them receive recognition.