How AI Is Rewriting the Economics of Security

Before You Dive In:

• Traditional security relies on an economic asymmetry where the cost of an attack outweighs the potential reward.
• AI is rapidly reducing the cost of vulnerability research and exploitation, turning once-secure systems into easy targets.
• Establishing true digital ownership now requires a transition to hardware-rooted security through hardware-backed solutions and cryptography.

Security Is an Economic Game

I often say there is no such thing as “100% secure.” People sometimes hear that as nihilism. It is the opposite. It is the most honest starting point for building real security.

Security wins when we create an asymmetry between defense and attack. The goal is straightforward: make attacks so difficult and so expensive that they are not worth attempting. If compromising a system would cost an attacker far more than the value they could extract, you are on the secure side. Equally, if the attack requires so much time that by the time the secret is reached, it is obsolete, the defender has won.

This asymmetry has been the foundation of security for decades. Everything we do, encryption, access control, patching, hardening, is ultimately in service of making the attacker’s job economically irrational.

That foundation is now cracking.

How Attackers Think: The Calculus of Opportunity Cost

To understand what is changing, you need to understand how attackers operate. They are rational economic actors. They think exclusively through the lens of opportunity cost: which target requires the least effort and yields the most profit?

Here is the critical nuance. When you start attacking a system, you have no certainty about the actual effort it will take to compromise it. You might spend weeks and hit a dead end. The cost is uncertain. But the reward side of the equation is remarkably clear. There is an established market, with actual brokers, for vulnerabilities and exploits. An attacker has a very good idea of what a working exploit is worth before investing a single hour.

This information asymmetry has historically benefited defenders (a bold claim, I know, given the endless stream of breaches scrolling through your news feed. Let me qualify: it benefited the defenders who invested seriously in security. The rest were already losing).

The uncertainty of effort acts as a deterrent. The attacker does not know if the target is a week of work or a year. That uncertainty, multiplied across all possible targets, has kept the security equilibrium roughly intact.

Until now.

The Vulnerability Market

The security ecosystem has matured into a full-fledged marketplace. On one side, you have skilled security researchers who discover exploitable vulnerabilities. On the other, you have buyers and brokers who connect the two.

The buyers fall into distinct categories. Vendors themselves purchase vulnerabilities to patch their products quickly. But more often, the buyers are nation-states, intelligence agencies, or criminal organizations willing to pay a premium for weaponizable exploits. The brokers — companies like Zerodium and Crowdfense — operate publicly, running acquisition programs with million-dollar payouts and functioning as market-makers with negotiation, escrow, and quality assurance services.

The prices tell the story. A zero-click iOS exploit chain now commands $5 to $7 million; Android exploits fetch up to $5 million. Chrome and Safari chains sell for $3 to $3.5 million, a 44% annual increase. The broader commercial surveillance software market is valued at over $16 billion, with over 435 entities across 42 countries participating in what has become a structured global industry.

In a perverse way, these high prices have kept a certain balance. They told us that breaking well-defended systems was hard, rare, and expensive.

That signal is about to become obsolete.

AI Collapses the Cost of Attack

The current equilibrium is being disrupted at its core. AI is dramatically reducing the cost of vulnerability research and, even more critically, the cost of exploitation.

What once required months of expert analysis, reverse engineering binaries, fuzzing at scale, and crafting reliable exploits is becoming faster, cheaper, and more accessible. AI-powered tools can analyze vast codebases in hours. They can identify vulnerability patterns that would take a talented security researcher weeks to spot. And they are improving at an extraordinary pace.

Here is how to think about what this means. In security, “secure” effectively means zero exploitable vulnerabilities, a perfect score, 20 out of 20. “Insecure” means having even one. For years, most well-defended systems operated at 19 out of 20 (optimistically): not perfect, but the remaining vulnerability was so expensive to find and exploit that it did not matter in practice.

The cost of finding that last flaw was the real defense. AI is removing that cost. Your 19 out of 20, which used to be effectively secure because exploiting it required months of expert work and a seven-figure budget, is now a zero. The vulnerability still exists, and now finding and exploiting it is cheap.

This is the shift. Services that were considered “secure enough” because attacks were uneconomical are now exposed. The economic equation that protected them no longer holds.

The Evidence Is Already Here

This is not speculation. The data is staring us in the face.

According to the ITRC’s 2025 Annual Data Breach Report, the number of data compromises hit a new record of 3,322 in 2025 — a 79% increase over five years. In 2024, over 16.8 billion records were exposed globally. This is not a fluctuation. It is a structural surge. Cyber-criminal activities that were once the domain of sophisticated groups are being commoditized. Attack toolkits are cheaper, more automated, and require less expertise to deploy.

The examples of AI-assisted attacks at scale are already legion. In early 2026, a solo operator used Anthropic’s Claude to breach multiple Mexican government agencies, exploiting at least 20 vulnerabilities and exfiltrating 150 gigabytes of sensitive data, tax records, voter registration files, and government credentials.

No custom malware. No command-and-control infrastructure. No nation-state backing. Just a commercial AI subscription and persistence. Operations that would typically take a coordinated red team two to four weeks were completed in under 72 hours.

Anthropic itself disclosed a Chinese state-sponsored campaign that used Claude Code to target roughly 30 global entities, tech companies, financial institutions, and government agencies, with AI autonomously executing 80 to 90% of tactical operations.

Perhaps the most alarming signal: general-purpose operating system exploits, the kind that were previously the exclusive arsenal of three-letter intelligence agencies, are now being used at scale by criminal organizations. Capabilities that cost tens of millions to develop a few years ago are trickling down the value chain at an accelerating rate.

🚨Only days after Coruna, one of the first large-scale iOS exploit kits, DarkSword is already being exploited in the wild.

Coruna showed the pattern: state-grade iOS exploits don’t stay in government hands. They leak, spread, and end up in broader ecosystems. One visit to a…

— Charles Guillemet (@P3b7_) March 18, 2026

And then there is the other side of the equation: the attack surface is exploding. AI allows anyone to generate software at near-zero cost, without deep technical expertise and certainly without any understanding of secure development practices. The volume of code being produced is skyrocketing, and a vast proportion of it is written or prompted by people who have no idea how to build secure software. Every line of that code is a potential entry point.

A World of Abundance, and Vulnerability

Let me state this plainly. We are entering a world of software abundance. AI generates code at unprecedented speed and at a fraction of the historical cost. Simultaneously, AI is extraordinarily powerful at finding vulnerabilities and exploiting them, especially when wielded by knowledgeable security researchers or motivated attackers.

The asymmetry that we, as a security community, have spent decades building is falling apart.

This is the fundamental shift. It is not just that attacks are getting cheaper. It is that the entire landscape is tilting. More software, more attack surface, lower cost of exploitation, easier access to advanced techniques. Each of these trends alone would be concerning. Together, they represent a phase change.

• Read ‘Revenge of the Atoms‘ by Ledger, CEO, Pascal Gauthier, exploring how blockchain and AI intelligence are colliding to fundamentally redefine the concept of trust.

Old Doctrines Won’t Save Us

For years, the security community has relied on a set of well-established principles: defense in depth, hardening, obfuscation, and security by obscurity. These approaches worked when the cost of bypassing them was high enough to deter most attackers.

AI renders these doctrines insufficient. Obfuscation does not survive an AI that can deobfuscate code in seconds. Security by obscurity does not hold when an AI can explore every possible code path exhaustively. Hardening helps, but when the cost of probing is reduced by orders of magnitude, it becomes a speed bump rather than a wall.

The bar for what constitutes real security is about to be raised dramatically. Recreating the defense-attack asymmetry, which is the only thing that actually makes systems secure, will require fundamentally different approaches. The principles that got us here will not get us there.

The Path Forward: Cryptography as the Foundation

Here is the good news. The tools that can restore the asymmetry already exist. We have known about them for years. What has changed is that AI is making them more tractable to implement at scale.

The common thread is cryptography. Not as a feature you bolt on at the end, but as the architectural foundation. Provable code. Provable execution. And when secrets must be protected at all costs, secure enclaves are the final bastion, a physical boundary that no amount of software intelligence can cross.

Secure Enclaves: Confidentiality and Secure Execution.

Hardware-rooted trust, physically enforced secure enclaves, such as Secure Elements, will become mandatory for any critical application. A secure enclave provides two things that software alone never can: confidentiality of secrets and integrity of code execution, even when the surrounding system is compromised. You cannot prompt-inject a silicon chip. You cannot social-engineer a hardware gate. When trust is anchored in physics, the attacker is fighting a fundamentally different battle. For critical operations, key management, identity, and transaction signing, secure enclaves are not an option. They are the floor.

Zero-Knowledge Proofs: Execution Integrity Without Trust Assumptions.

ZK technology gives us something remarkable: the ability to prove that a specific computation was performed correctly, with specific inputs, producing specific outputs, without revealing the internals and without requiring trust in the party that executed it. This is not about hiding information. It is about providing cryptographic certainty that your critical code ran as intended, with no trust assumption on the executor. For systems where the integrity of execution is paramount, such as financial settlement, smart contracts, and critical infrastructure, ZK proofs offer a level of assurance that no amount of testing, auditing, or monitoring can match.

Formal Verification: Provable Code.

Mathematically proving that code behaves exactly as specified, for all possible inputs, in all possible states. This has been the gold standard of software correctness for decades, and for decades, it was largely intractable for real-world systems. The proofs were too complex, the tools too slow, the expertise too rare. AI is changing this equation dramatically. AI-assisted formal verification is making it possible to prove properties of code at a scale and speed that was unimaginable five years ago. Critical code, the code that protects secrets, manages keys, and executes transactions, should be formally proven. What was once a luxury reserved for avionics and nuclear systems is becoming a necessity for any software that handles value.

The Holy Grail: FHE and Indistinguishability Obfuscation.

Looking further ahead, Fully Homomorphic Encryption, the ability to compute on encrypted data without ever decrypting it, and Indistinguishability Obfuscation, making code truly uninspectable even to someone with full access to it, would represent the ultimate security primitives. FHE would eliminate the need to ever expose data during processing. iO would make reverse engineering mathematically impossible. Together, they would close the gap entirely. But let me be honest: even with AI, these technologies remain intractable today. The theoretical breakthroughs are real, but practical deployment at scale is still (optimistically) years away. They are the destination, not today’s toolkit. For now.

Let me be equally direct about the difficulty ahead. AI is making these tools more accessible, but the operational, financial, and skills barriers to deploying them at scale remain enormous. Rewriting critical infrastructure around secure enclaves, ZK proofs, and formal verification will take time, significant investment, and expertise that is still scarce. This is not a quick fix. It is a generational effort, one that needs to start now.

The Operational Reality

Not every system can be rewritten from scratch with formal proofs, wrapped in a secure enclave, and verified with ZK. The real world is messy. Legacy systems exist. Budgets are finite. Time is short.

In practice, the security community will continue to follow a layered approach. Supply-chain integrity, knowing exactly what code you are running and where it came from, will be a critical first line of defense. Targeted rewrites of the most exposed components into memory-safe languages will eliminate entire classes of vulnerabilities at their root. And for the vast majority of systems that cannot be perfectly hardened or mathematically proven, the ability to detect compromise quickly, patch at speed, and contain the blast radius will be among the few scalable ways to reduce damage.

What increasingly matters is speed. The time between vulnerability disclosure and real-world exploitation of known vulnerabilities (1-day / n-day) is shrinking rapidly, leaving defenders with an ever smaller window to respond. AI is already being deployed defensively: automated code scanning (SAST, DAST) that catches vulnerability patterns before code ships, behavioral detection systems (EDR, XDR, UEBA) that surface attacker activity in real time, and patch automation that shrinks the window between disclosure and fix from weeks to hours. These are not theoretical; they are operational today, and they represent the immediate line of defense while the deeper cryptographic foundations are built.

The same technology that is lowering the cost of attack is also lowering the cost of defense. The first question is who moves faster. The second one is, does it recreate the asymmetry?

Let me be blunt: it does not.

The Race Is On

The old equilibrium is gone. The economic assumptions that underpinned “secure enough” have been invalidated by AI. Every system, every service, every application that was designed without security as a first principle is now living on borrowed time.

But this is not a story of inevitable defeat. It is a story of urgency. The tools to build genuinely secure systems, secure enclaves, zero-knowledge proofs, and formal verification are more powerful and more accessible than ever. AI is already strengthening defenses through code scanning, anomaly detection, and automated patching. And for the systems that cannot yet reach the highest bar, practical measures, supply-chain integrity, memory-safe rewrites, rapid detection, and containment can meaningfully narrow the gap.

The question is not whether these tools will be adopted. They will. The question is whether we adopt them before or after the wave of breaches forces our hand. The security community, the software industry, and every organization that holds sensitive data or manages critical infrastructure must move now.

The asymmetry can be rebuilt. But the window to do it proactively is closing fast…

---

Authored by Ledger CTO, Charles Guillemet