Two Giants Bet Big on AI Defense
The global arms race in artificial intelligence is taking two dramatically different forms. In the United States, defense startup Anduril has secured a landmark contract worth up to $20 billion from the US Army — the largest single enterprise contract in the military's history. Meanwhile in China, 360 has unveiled its "Security Lobster" series, a new product line designed to protect AI agents using a model-against-model approach.
Anduril's Historic $20B Deal
The US Army announced on March 14th that it has awarded Anduril a contract valued at up to $20 billion, consolidating over 120 separate procurement projects into one unified agreement. This massive consolidation reflects the Pentagon's push to modernize its defense infrastructure around AI and autonomous systems.
Anduril, founded by Oculus VR co-founder Palmer Luckey, has positioned itself as a key player in defense AI. The company's previous projects include autonomous drones, AI-powered surveillance towers, and counter-drone systems. This contract signals that the US military is betting heavily on AI to maintain its technological edge.
The deal comes amid growing concerns about China's rapid advancement in military AI. Recent reports indicate that Chinese defense contractors are accelerating their own AI development programs, creating what some analysts call a new kind of arms race focused on intelligent systems rather than traditional weaponry.
360's Model-Against-Model Security
On the same day, Chinese internet security giant 360 announced its "Security Lobster" (安全龙虾系列) product line — a novel approach to securing AI agents through adversarial testing. The "model-against-model" methodology essentially pits AI systems against each other to identify vulnerabilities before malicious actors can exploit them.
The Security Lobster series represents 360's answer to a growing problem: as AI agents become more autonomous and integrated into business workflows, they become attractive targets for hackers. Traditional security tools designed for static software struggle to cope with dynamic, learning systems that can behave unpredictably.
By simulating attacks between AI models, 360 claims it can proactively discover weaknesses in agentic AI systems. The approach mirrors how red teams operate in cybersecurity, but replaces human penetration testers with AI models that can test thousands of attack vectors in minutes.
What This Means for the Industry
Together, these announcements underscore a fundamental shift in how both governments and technology companies approach defense and security. The days of building walls around digital assets are giving way to a new paradigm: intelligent systems defending against intelligent threats.
For enterprises deploying AI agents — whether for customer service, code generation, or data analysis — the message is clear: agent security is no longer optional. As 360's lobsters demonstrate, the next generation of cybersecurity will be fought by machines, for machines.
Anduril's windfall suggests that governments are equally willing to pay premium prices for AI-powered defense capabilities. With $20 billion on the table, the race for military AI supremacy is clearly just beginning.