Today’s cybersecurity systems are not ready for AI

The digital economy has led to a U.S. trade surplus in services on the order of around 1 percent of gross domestic product. Yet recent advances in artificial intelligence have exposed new vulnerabilities that place every cyber system at risk of disruption.

Cryptography has been crucial to protecting cyber systems. Using hashing or encryption, passwords can be protected, which has historically kept cyber systems safe from intruders. More recently, quantum computing has been viewed as a threat to shatter cyber system security, given its ability to unlock such systems in a reasonable amount of computing time. This has pushed the discovery of quantum-ready cryptography — cryptography that even quantum computers cannot break.

This traditional model for protecting cyber systems is akin to installing an impenetrable lock on a house door to keep criminals from entering. AI systems that now exist effectively allow such people to enter the house through other means, without touching the house door that had become the focal point of protection.

Enter Claude Mythos, a set of large language models created by Anthropic that can uncover cyber system security weak spots and even provide recommendations for fixes. Such vulnerabilities can let password protections be circumvented, or structural components of software code be identified that allows bad actors to sidestep cybersecurity protections, accessing data and other proprietary information.

To date, Anthropic has not shared Claude Mythos publicly, but rather with large digital companies like Microsoft, Apple and Amazon, whose business models rely on robust cybersecurity.

This was very responsible of Anthropic, but it does not change the underlying reality: The mere fact that AI systems have been created to penetrate cyber systems and circumvent security protections means that such systems will eventually be created by less honorable people and entities and used for nefarious purposes. This is why cyber security stocks initially dropped after the capabilities of Claude Mythos were announced.

What is becoming clear is that AI risks now include the entire digital economy. Critical infrastructures like financial services, energy, health care and public health are all potential targets to be compromised. The number of security breaches over the last decade has skyrocketed. With AI systems now positioned to facilitate cybercrime, every system has become a vulnerable target.

Just as computers need antivirus software to protect against external attacks, cyber systems will need protection against AI attacks. More importantly, the design of cyber systems will need to be rethought, to consider how AI may be used to exploit vulnerabilities. Although an ideal of zero vulnerabilities is likely unachievable, working toward such an objective may be the only way to stay ahead of the AI risks.

Although the digital economy has proven itself to be remarkably robust, new AI developments make it increasingly fragile and vulnerable to disruption. Though Claude Mythos may not be publicly available at this time, its capabilities will eventually be reproduced in some form and could end up in the wrong hands. It’s not a matter of if, but of when.

Claude Mythos demonstrates why regulation of AI cannot solve this problem. It also highlights why preparations for such an inevitable future must begin today. Creating multiple levels of independent cyber barriers is a natural rudimentary first step, with the costs ultimately borne by consumers. Still, designing AI-resistant software will be the ideal, effectively creating a new arm in the cybersecurity industry.

Given the speed at which Claude Mythos was created, Anthropic did the right thing to share its abilities with the largest software companies as plans for a new digital economy reality begin to take hold.

Sheldon H. Jacobson, Ph.D., is a professor of Computer Science at the University of Illinois Urbana-Champaign. He applies his expertise in data-driven risk-based decision-making to evaluate and inform public policy.

More Opinions - Cybersecurity News

Republicans fear succession of government shutdowns under Trump

Fetterman on Democrats’ redistricting win in Virginia: ‘We all lose at this ...

Supreme Court leak shakes institution’s secrecy

Iowa Republican zeroes in on birth control access in heated Senate race

Spotlight shifts to Florida after Democrats win Virginia redistricting battle

Patel gets in shouting match with reporter as he defends job performance

Cruz: Schumer will shut down government weeks before midterms

Walz: ‘Next Democratic president better figure out a way to get universal ...

Finally, MAGA figured out who the real Donald Trump is

Watchdog requests Patel’s calendar, security detail messages

The Memo: Can Trump realistically better Obama’s Iran deal?

Jeffries: Democrats working to ‘make sure’ Hegseth, Patel ‘are on their ...

Cuban says ‘no’ when asked if he wants Harris to run for president in 2028

Trump rips Carville over call for more states, Supreme Court justices

Trump extends Iran ceasefire indefinitely, in an about-face

Mills-Mace feud blows up into tit-for-tat expulsion threats

Democrats boost midterm edge with Virginia redistricting win: 5 takeaways

WHCA buzz reaches fever pitch as Trump set to attend for first time as president

The Hill Podcasts – Morning Report

© The Hill

visit website