Six months after Mythos, here’s how AI is shaking up vulnerability management: Experts
While a massive overhaul is underway for the process of finding and fixing vulnerabilities in the wake of Anthropic’s Claude Mythos being revealed, the progress has been far from even around the industry, solution provider execs and security experts tell CRN
The arrival of ultra-powerful AI models for vulnerability discovery that began with Anthropic’s Claude Mythos earlier this year has jolted the industry into making widespread changes in exposure management practices in just six months, though the challenges remain steep for many organisations, cybersecurity experts tell CRN.
The 7 April disclosure of details about Mythos’ proficiency in discovering and exploiting software vulnerabilities at unprecedented speeds led to an immediate mobilisation within many businesses around the issue.
Given that many companies can take weeks or months to patch vulnerabilities after disclosure, the prospect of a monumental surge in new software bugs was met with attention, and even panic, at the highest levels of industry and government.
Indeed, the six months since Mythos came to light, which has also included a cascade of other frontier AI advancements in the area, including from OpenAI and open-weight models, have seen massive changes sweep the industry around vulnerability and exposure management.
Still, when it comes to the processes for finding and fixing vulnerabilities, progress has been far from even across the industry, solution provider executives and security experts tell CRN.
Without a doubt, the arrival of Mythos and other frontier models for vulnerability discovery “has really created an opportunity for our clients to push their threat exposure management agenda into the board, because boardrooms are now listening,” says Kathryn Hall, SVP of services at Kansas-based Optiv.
The result has been “a lot of really meaningful momentum” in a relatively short span of time, Hall says, though there’s no question that the degree of progress “has significantly varied by organisation”.
Key changes that have come to the industry include a notable shift in how businesses decide which vulnerabilities to fix and how continuously they assess their exposure, as well as an evolution in how leaders make the case for expanded security investments to the board, according to experts.
At the same time, as new capabilities boost visibility into vulnerabilities and help identify the most important exposures to address, many businesses are still running into the same familiar obstacles that get in the way of rapidly fixing vulnerable software.
A shift in the conversation
For consulting giant Accenture, there’s no question that the typical customer conversation has changed dramatically since the initial alarm over AI-powered vulnerability discovery sparked by Mythos, according to Accenture’s Jason Lewkowicz.
Initially, calls from concerned clients largely centred around how to patch vulnerabilities more rapidly as a way to address the fact that the models were identifying software flaws so much faster, says Lewkowicz, global lead for cyber resiliency and defence at Dublin-based Accenture.
“I was finding more clients spending time on wanting to patch faster, versus rethink their broad-stroke programme,” he says.
“In the beginning, it was probably 80 per cent, ‘Let’s go patch faster and play whack-a-mole.’”
However, as time went on, and with the help of Accenture, many customers eventually came to see the limitations of that approach and recognise the need for a broader update to their strategy, Lewkowicz notes.
“Now it’s probably 80 per cent in the bucket of, ‘Let’s really rethink our approach to this,’” he says.
That is translating, for instance, into heightened interest among customers in methods such as continuous threat exposure management (CTEM) and attack surface identification, as well as penetration testing and offensive security, Lewkowicz says.
Without a doubt, Mythos has accelerated a transition towards CTEM that was already underway, according to GuidePoint Security’s Gary Brickhouse.
“Mythos is fast tracking it, frankly, and organisations probably are having to deal with it at a faster clip than they were originally planning to,” says Brickhouse, CISO at Virginia-based GuidePoint Security.
Previously, “I think maybe everybody felt like, ‘Hey, we can sort of ease into this,’” he says.
The truth is that CTEM is more of an ongoing approach to understanding and reducing exposure, as opposed to a particular goalpost that an organisation reaches, Brickhouse says.
The idea is for businesses to be continually assessing their attack surface and identifying new routes an attacker could take.
“Probably the biggest, most important piece of CTEM is the ‘C’, which is continuous,” Brickhouse says.
The ultimate purpose is to give businesses an ongoing process for pinpointing which vulnerabilities and exposures truly matter most, according to Brickhouse.
Six months after Mythos, the advancement toward better prioritisation of exposures is “where we see most of the maturity happening,” he says.
Chaining of vulnerabilities
Another critical aspect of responding to the frontier AI shift has been the recognition that these models bring a proficiency for connecting vulnerabilities together that simply was not possible in the past, experts tell CRN.
“The reality is, yes, AI is being used to find new vulnerabilities,” says Rocky Giglio, founder and CEO of South Carolina-based Cloud Security Pros.
“But simultaneously, AI is also being used to chain vulnerabilities together.”
That has been highly relevant for businesses looking to adapt their vulnerability management programmes for the Mythos era, according to Giglio.
The reason, he says, is simple: fixing only the vulnerabilities with the highest severity scores can leave a backlog of bugs that threat actors may now be able to chain together to carry out cyberattacks.
Addressing that challenge requires a much more complete view of the potential attack path, and AI can help defenders with that too, Giglio says.
For instance, an organisation may have a vulnerability on a web server that is already protected by a web application firewall rule, whereas another bug on a virtual machine with a public IP address could actually warrant the most immediate attention.
New AI-powered capabilities can make that distinction very effectively, he says.
The idea is “applying AI to the whole attack path rather than just, ‘You have this vulnerability on this system,’” Giglio says.
Faster patching
Alongside deploying new capabilities for prioritising and fixing vulnerabilities, many businesses still need to update their practices around vulnerability management to enable faster responses, Giglio notes.
Without a doubt, in the Mythos era, “you can’t have a monthly patch cycle,” Giglio says.
“That’s too long.”
Many customers are, in fact, revisiting restrictions that in the past allowed potentially disruptive software updates to take place only once a month, according to GuidePoint’s Brickhouse.
Even with the adoption of new patching policies, however, many businesses still need to make difficult decisions about where to direct their limited resources, he says.
New AI capabilities can help in that regard as well. The approach involves using AI to connect signals from different security tools, with the context necessary to determine which exposures are the highest priority for the business, Brickhouse says.
One way that Troy, Michigan-based Logicalis is working with clients to accelerate patching processes is to agree up front on an expedited “emergency change process”, says Troy Saunders, CISO at Logicalis US.
“It’s about setting up all the parameters and [having an] agreement up front around what can we go ahead and patch without approval,” Saunders says.
That process can enable the solution provider to establish which changes can proceed without another round of approvals while ensuring that actions are documented, he notes.
Communication is key
At cybersecurity powerhouse Optiv, many customer engagements are demonstrating both the progress being made and the importance of addressing key organisational issues surrounding remediation processes, according to Hall.
For one manufacturing customer, for instance, Optiv developed a new vulnerability management programme and implemented the tools needed to improve visibility, she says.
But Optiv’s largest contribution has actually been around helping different business units communicate and take responsibility for addressing exposures.
“This is not just a security problem,” Hall says.
“This touches every business unit and infrastructure, and it takes a lot of stakeholder communication, education and change management.”
Uneven progress
Still, greater awareness, and helpful new capabilities leveraging AI, have not universally translated into dramatic improvements in vulnerability management, according to security experts.
“I think people have been very interested in moving [to a new approach],” says Rob Lefferts, corporate VP for threat protection at Microsoft.
“I’m not sure they are moving fast enough in practice yet. And that's especially true for the organisations that don't have a strong security focus.”
Another issue is that the constant arrival of new tools and competing claims can actually make it harder to decide what to do next, says Giglio.
“There’s a little bit of the analysis paralysis. There are so many choices,” he says.
“There's a new cybersecurity tool every day that's promising to solve some element of this.”
Meanwhile, although customers are becoming more willing to fund the investments needed to update vulnerability and exposure management programmes in the wake of Mythos, that investment does not appear to be happening fast enough across the board, according to Hall.
And in any case, funding alone will not entirely resolve the challenges posed by AI-accelerated vulnerability discovery, she notes.
“That money can buy you some tooling, but someone has to implement it and run it, and still drive all that stakeholder communication and connectivity,” Hall says.
The bottom line is that while heightened attention from boards and C-level executives has provided greater support for making necessary security changes, turning that support into lasting improvement remains a work in progress for many businesses.
“Yes, I’m optimistic,” Hall says, “but there’s a lot of work to be done.”
This article originally appeared on CRN UK sister websiteCRN.