With the rapid rise in the use of AI, large language models (LLMs), and retrieval-augmented generation (RAG), the IT security landscape is undergoing a seismic shift. Recently, Anthropic previewed a new foundational model, Claude Mythos, that has shaken the cybersecurity community.
Although Mythos is a general-purpose model, it has demonstrated a significant ability to handle computer security tasks. In a blog post, Anthropic states, “During our testing, we found that Mythos Preview is capable of identifying and then exploiting zero-day vulnerabilities in every major operating system and every major web browser when directed by a user to do so. The vulnerabilities it finds are often subtle or difficult to detect.” It is also important to realize, certainly in the context of Java and its 30-year history, that many of the vulnerabilities detected have remained undetected for years, even decades. The oldest found so far is a 27-year-old bug in the OpenBSD operating system.
What is even more concerning for security teams is the level of sophistication employed by Mythos. In one example, Mythos constructed a web browser exploit that chained together four vulnerabilities. Each of those vulnerabilities used in isolation may not have resulted in compromised software, but the holistic effect became meaningful.
Claude Mythos is not the end of this story; in many ways, it is just the beginning. Anthropic has developed an LLM that can now find and exploit vulnerabilities, outperforming both human experts and existing tools.
The security problem will be further exacerbated by the rise in vibe coding and the broader use of AI coding tools like Cursor, GitHub Copilot, and Claude Code. Organizations will be relying on opaque AI-generated code. As more of the software development process, such as testing, uses AI, we will have to place increasing trust in these tools.
Anthropic is being very responsible, given Mythos’s potential impact, and has not yet made the model publicly available. Under the recently announced Project Glasswing, only a select group of companies have been granted access to Mythos. The goal is to enable providers of critical software infrastructure to use Mythos as part of their defensive security work. This will cover operating systems, web browsers, and, most likely, Java.
When successful, cyber hacks can drastically impact even the most well-known and successful companies. The NotPetya malware used stolen NSA hacking tools and is estimated to have cost Maersk, the shipping company, $10 billion in damages. The Yahoo data breach, the largest in history, exposed three billion user account details. The Equifax data breach exploited a known, unpatched vulnerability in a web application framework. The highly sensitive personal data of nearly 150 million consumers was compromised. Realistically, this is the tip of the iceberg. Many organizations will not reveal publicly that they’ve had their IT systems compromised for fear of reputational damage.
What we’ve seen is a continual battle between cybercriminals and security teams, each trying to find vulnerabilities more quickly than the other. When located, the contest switches to whether the vulnerability can be exploited before it is patched. Each side has employed increasingly sophisticated automation. Cybercriminals use tools like port scanners, password crackers, and web application scanners. Security teams use commercial tools such as Black Duck and Snyk to scan source code for vulnerabilities.
Open-source software is now the foundational infrastructure of modern technology. The source code for everything from Linux to Firefox to Kubernetes to Java is freely available for anyone to download. Cybercriminals no longer need to guess how software works; they can just look at the source code and try to outsmart the security scanning tools. AI amplifies this advantage, analyzing entire codebases at machine speed and finding the subtle flaws that automated scanners routinely miss.
As a fundamental component of most enterprise and mission-critical applications, the Java platform has an extremely rigorous and well-structured security strategy. Within the OpenJDK open-source project, there is a team known as the OpenJDK Vulnerability Group (OVG). Despite the project’s open nature, this team operates under a cloak of complete privacy. Membership of the group is carefully controlled, and only engineers with the requisite security background and skills are selected. Since patches are being developed for currently unpatched vulnerabilities (those that would be classified in the open as zero-day), all communication is encrypted. Unlike other groups in the project, this group’s email exchanges are not published.
Based on the OVG’s work, a well-defined schedule delivers source code changes for security patches on the third Tuesday of January, April, July, and October. Most (but not all) of the major OpenJDK binary distribution providers have representatives on the OVG. This ensures that updates to Java runtime distributions are made available to users. As the principal developer of Java, Oracle releases its patches as a Critical Patch Update (CPU) at the same time the source code repository is updated. Once Oracle’s CPU is publicly available, the embargo on other OpenJDK distributions is lifted, and they are free to publish their update. How long this takes depends on the provider; obviously, the longer the delay, the greater the risk that unpatched vulnerabilities will be exploited. For Azul Core, which includes the Zulu builds of OpenJDK, updates have consistently been made available within one hour of the embargo being lifted.
To ensure the highest level of security for your JVM-based applications, you should have a well-defined strategy for deploying JDK updates across your estate. It is all too common to hear users say they do not need to patch their machines because they are behind a firewall. A firewall is not impenetrable. The Equifax breach mentioned earlier used an exploit on machines outside the firewall; once those machines were compromised, the attackers could pass through the firewall as if it weren’t there.
Part of this strategy should be a clear inventory of which machines in your estate are running which Java versions, and even which specific updates are deployed.
When Java updates are released, there should be a process for evaluating the vulnerabilities being addressed so that a rapid decision can be made about how quickly the update should be rolled out. To ensure both maximum security and stability for applications, JDK updates can be provided in two forms: the CPU, which includes only security patches and the Patch Set Update (PSU) or full update, which includes all changes. Not all distributions provide a separate CPU (Azul is one of only two that do). Given the frequency of updates, rigorous testing with all popular applications, libraries, and frameworks is not possible. Roughly one in four PSUs contains a fix that breaks something else, requiring an update to the update. CPUs have never been affected in this way, so they can be used to deploy critical patches without affecting stability.
As I mentioned, this is only the start of the AI cybersecurity story. Other frontier model developers will start delivering models that can match and exceed Claude Mythos’ exploit discovery capabilities. We are entering a new and more perilous time for cybersecurity. To that end, users must be more vigilant than ever about keeping their Java runtimes up-to-date.
—
New Tech Forum** provides a venue for technology leaders—including vendors and other outside contributors—to explore and discuss emerging enterprise technology in unprecedented depth and breadth. The selection is subjective, based on our pick of the technologies we believe to be important and of greatest interest to InfoWorld readers. InfoWorld does not accept marketing collateral for publication and reserves the right to edit all contributed content. Send all inquiries to *** doug_dineley@foundryco.com.*