Could AI push governments back to demanding digital backdoors?

For years, law enforcement has operated under an uneasy truce. Rather than forcing tech companies to build permanent access points into devices—a move that would weaken security for all users—governments have increasingly relied on purchasing private hacking tools and zero-day exploits. This market-based approach allows authorities to bypass encryption on a case-by-case basis while maintaining the illusion of a secure digital ecosystem.

However, this equilibrium faces a new threat: AI-driven security. If Large Language Models become efficient enough to identify and patch bugs at scale, the supply of exploitable vulnerabilities could dry up. Paolo Stagno, CTO at the exploit-brokerage firm Crowdfense, notes that while the current system of buying vulnerabilities is the most democratic model available, the status quo is fragile. If the "gold rush" of discoverable bugs ends, state agencies may find themselves unable to maintain their surveillance capabilities.

Not all experts agree that a total drought is imminent. Some, like DarkCell founder Hamid Kashfi, argue that AI will simply widen the gap between reported bugs and the complex, undiscovered flaws that remain valuable to intelligence services. Meanwhile, others point out that modern security protections and the sheer complexity of current software are more significant hurdles to hacking than the threat of AI patching. Still, the underlying pressure remains: as long as authoritarian regimes demand exceptional access, the temptation to mandate backdoors may eventually override the current preference for market-based espionage.