Anthropic has an unreleased model that enabled a research team to break through the M5 chip hardware defenses in just five days—work that Apple invested five years and billions of dollars to build.

This is not an ordinary vulnerability discovery. The Calif team used the Mythos Preview model to build the first publicly available kernel memory corruption exploit chain from scratch on a Mac equipped with the M5, directly bypassing Apple’s latest MIE (Memory Integrity Enforcement) mechanism. The key point is: Mythos is good at generalizing known vulnerability types, quickly pinpointing flaws, and then human experts complete breakthroughs against new hardware defenses.
This combination—AI-assisted discovery plus targeted human exploitation—is rewriting the efficiency curve of security offense and defense. What used to take months of reverse engineering can now be compressed into five days. Apple’s hardware security team may face unprecedented pressure: traditional trust models based on hardware isolation are becoming blurred in the face of AI-assisted vulnerability hunting.
For the crypto industry, this is not distant AI news. The security architecture of DeFi protocols, cross-chain bridges, and smart contract platforms fundamentally relies on similar hardware/software trust assumptions. If Apple’s hardware defenses—backed by billions of dollars—can be breached in five days by an AI-assisted team, then on-chain protocols with weaker security budgets may have a much larger risk exposure than previously expected.
The counter-risk is that the narrative may overstate AI’s attack capabilities while ignoring the defenders’ countermeasures—security teams can also use AI to accelerate patch development and vulnerability remediation. Both offense and defense are moving faster, but in the short term attackers may hold an asymmetrical advantage, especially when targeting long-unpatched legacy systems.
This isn’t a reason to panic, but it should be a signal to reassess how security budgets are allocated: the efficiency and cost of AI-assisted attacks are decreasing, and protocols’ security audits and monitoring need to be upgraded accordingly.
$ai #defi # On-chain data #ai # Blockchain