Iran-backed Houthi fighters captured the strategic Yemeni port city of Mocha on Thursday after deploying an artificial intelligence-generated audio clone of Lieutenant General Tareq Saleh to trick his troops into retreating.

In what might be history’s first military victory driven by high-tech deepfakes and combat sandals, the insurgent group managed to fool the UAE-backed National Resistance Forces into thinking their top commander was ordering a full surrender. The cloned audio file, which spread rapidly across social media via Yemeni activist Adel al-Hasani, copied Saleh’s voice with convincing precision, instructing officers to abandon critical positions near the strategic Bab al-Mandeb Strait.

The psychological operation extended beyond synthetic voice memos. At around 2:00 AM on September 10, Houthi forces dispatched targeted text messages to soldiers along the western coast, offering full amnesty to anyone willing to drop their weapons and walk away. Combined with the commander’s fake audio command echoing on their phones, the defensive line crumbled, allowing Houthi fighters to march into Mocha and claim their largest territorial prize since the 2022 UN-brokered truce.

The fabricated recording painted a dramatic picture of battlefield collapse that completely bamboozled troops on the ground:

"After assessing the situation, and following betrayal from within, the failure of our allies and heavy losses suffered for the homeland and the Republic, all units must withdraw from Mocha so that the enemy can be targeted in open terrain... I repeat, all units must withdraw to Mocha so that the enemy can be targeted in open terrain."

Mocha-based satellite channel Aljoumhouriya TV scrambled to issue a denial, posting on X that the broadcast was merely a "failed attempt to mislead public opinion and affect morale." However, the debunking arrived far too late to prevent confused soldiers from taking the bait. Saleh’s military media team subsequently warned troops that any future commands released outside official channels should be considered fake.

Adding an extra layer of tech-savviness to the incident, AI safety firm Anthropic released a 154-page threat intelligence report on Thursday confirming that a weapons group in Houthi-controlled northern Yemen had also experimented with its Claude AI model. The insurgents reportedly leveraged the software to write code for guided rockets, multi-stage ballistic missiles, and hypersonic gliders, proving that low-tech frontline fighters can manipulate cutting-edge artificial intelligence with surprising effectiveness.