A security research team just used Claude Mythos to identify the first known exploit in Apple’s M5 chip. They needed physical access to the device to use it, the vulnerability has since been patched, and I don’t think it should be seen as a huge threat. But it is a stark warning that in this AI age, attackers can find and exploit system vulnerabilities at a dangerously fast rate.
While widely reported, the proof-of-concept exploit was of limited significance because it required direct physical access to the target device; what matters most is that it is a very real illustration of the new security reality.
AI doesn’t care whose side you’re on
AI boosts productivity for everyone, including attackers. In this case, the technology augmented the human security research team’s efforts, enabling them to identify a weakness in Apple’s security system. This won’t be the first time AI gets used to identify hard-to-find bugs and certainly won’t be the last.
This should be a real concern to any platform provider, as it means the most well-resourced attackers will be leaning deep into AI to help them find vulnerabilities. And as AI improves, the capacity it provides will inevitably become more dangerous.
That’s even before you consider that some attackers work for the kind of state and state-adjacent entities that can afford aircraft carriers.
When nation-states come knocking
Access to such extensive resources means future AI-augmented attacks will have at their disposal the most powerful computational AI money can buy, which probably boils down to quantum computers.
The threat of quantum computers has been discussed since the 1990s. These systems are expected to be quite capable of breaking the encryption keys on which digital existence is built, and things will not be the same when they do. We don’t have long to wait until that threat becomes real. Google recently warned quantum computers will be able to hack into some, though not all, encrypted systems by 2029.
Once Q-Day breaks, there will be no going back. And just as Mythos AI was able to help security researchers break into Apple’s core security today, quantum computer-augmented AI is likely to open even more dangerous security frontiers. The Global Risk Institute warns organizations “should take immediate action to address this significant cyber risk.”
What you can do while the industry catches up
What actions can we take now? We have to look to the tech firms to develop tech to protect us against tech:
- Google, for example, is investing in post-quantum cryptography (PQC) digital signature protection and will put similar protections in place across its own authentication services in the next couple of years.
- Apple says it has also deployed quantum-secure cryptography across a wide range of protocols, “prioritizing applications involving sensitive user information where attackers could harvest encrypted communications at scale.”
- Cisco is deep into developing quantum-secure networks, working with network partners such as Orange Business to protect enterprise and public-sector data from future quantum threats.
These protections will help secure some of the most important elements of the computing experience, but they won’t cover everything, leaving a less certain threat environment in which many of the least-resourced software developers are exposed.
Legacy systems, particularly around critical infrastructure and key industries like health or finance, will be particularly exposed. You really, really don’t want key systems at your hospital or bank to rely on insecure and unsupported Windows 10 machines, for example. (You’d be better of with a MacBook Neo — truly, you would.)
Apple is not a badly resourced developer, which means it has no choice but to invest heavily in additional security to protect its platforms against both quantum and AI threats. We’ve also got to prepare for even more complex attacks down the road, as the two powerful technologies converge (to the detriment of security).
Gather ye while you can
We know nation-state actors are already hoovering up huge quantities of encrypted data, as they hope to be able to decrypt it once quantum capability matures. (There’s even a name for this, HNDL, which stands for Harvest Now, Decrypt Later.)
You don’t need to panic. These new breed attacks will be massively expensive to put together, which suggests they’ll be used against the same high-value targets Apple built Lockdown Mode to protect. What we saw happen with the NSO Group, which made the Pegasus attack you can now buy on the dark web, and other similar exploits that leaked over time, show that sophisticated attacks will inevitably seep into general use over time.
For now you can choose to use quantum ready messaging services such as iMessage and, while we wait for PQC-compliant password managers to ship, at least make sure to use highly secure passwords for key apps and services. And monitor the news for AI-augmented security exploits against Apple equipment. And as always, never leave your Mac unattended anywhere people you don’t know or trust can reach it.
You can follow me on social media! Join me on BlueSky, LinkedIn, and Mastodon.