Following the announcement of Mythos and Project Glasswing, institutions throughout the world are grappling with the potential dawn of a new era of cybersecurity. In this post, we break down the current situation, discuss the role of openness, and situate the future of cybersecurity within the larger AI ecosystem.
Mythos is a “frontier AI model”, a large language model (LLM) that can be used to process software code (among many other things). This follows a general trend in LLM development, where LLM performance on code-related tasks has recently skyrocketed. What’s particularly significant about Mythos is the system it’s embedded within: It's the system, not the model alone, that has enabled Mythos to rapidly find and patch software vulnerabilities. Understanding this distinction is key to understanding the current landscape of AI cybersecurity.
What Mythos demonstrates is that the following system recipe is powerful:
Together, these ingredients can uncover software vulnerabilities, find exploits, and build patches. It’s in this recipe — not in any one model — that both the benefits and the risks come in.
This matters because others can build comparable systems. Smaller models embedded in systems built with deep security expertise could potentially produce similar outcomes more cheaply, which is particularly promising for defense. AI cybersecurity capability is jagged: It doesn’t scale smoothly with model size or general benchmark performance. The system the model is embedded within matters a lot.
So what Mythos has demonstrated is that it’s possible to build an AI system that finds and addresses software vulnerabilities. We already knew this was possible and there has been increasing work on this, but we’re just beginning to explore what it means in the context of agentic AI: Systems that can rapidly and autonomously take action.
As autonomous systems that identify software vulnerabilities proliferate (and they will), open code and tooling can help level the playing field. Software security has become a speed race across four stages: detection, verification, coordination, and patch propagation. Open ecosystems distribute these across a community, where more closed-source projects centralize knowledge and action across all four stages inside a single vendor, representing a single point of failure where only one organization can see and fix the code. The distributed nature of open development is robust to such constraints, and can be especially powerful in communities with dedicated security professionals, like the Linux kernel security team, the Open Source Security Foundation, and the team at Hugging Face working on model and supply-chain security.
A common argument for more closed systems is proprietary obscurity, where the code underlying a system is inaccessible. Unfortunately, this provides less protection than it used to. AI systems are increasingly able to assist with reverse engineering of stripped binaries, which matters because most legacy firmware and embedded code is closed, binary-only, and no longer maintained. That code represents a huge attack surface, and it’s becoming more legible and accessible as AI tools improve.
There’s also a risk created by how AI is being used inside closed codebases. When companies adopt AI coding tools under the wrong incentives (for instance, evaluating engineers by the volume of features shipped rather than code quality) AI-accelerated development can introduce more vulnerabilities into proprietary code than traditional development would. Those vulnerabilities then sit inside a closed codebase where only one organization can find and fix them, while AI-enabled attackers are increasingly capable of discovering them from the outside. The combination of more vulnerabilities produced more quickly, behind a single-organization firewall, is exactly the kind of imbalance that open ecosystems are positioned to avoid.
Underlying all of this is capability asymmetry between attackers and defenders. Open models and open tooling narrow that gap by giving defenders access to the same class of capabilities attackers can reach for — capabilities that would otherwise be concentrated within a small number of well-resourced entities.
Cybersecurity defense is where open source and AI agents can play a key role together. Based on the System Card, it appears that Mythos is capable of operating with close to full autonomy, something we’ve advised against due to the potential loss of control. AI agents that are instead semi-autonomous, where the types of actions they can take are prespecified and certain steps require human approval, hit a sweet spot of benefit and risk. In semi-autonomous systems, people remain in control, and the AI agent is responsible for specific subtasks. This is possible to do with open code that organizations can run privately within their own institutions, specifying allowable tools, skills, and system access privileges. With this setup, AI agents can be deployed defensively, finding vulnerabilities and assisting with patching under an organization’s own controls.
The semi-autonomous approach depends on humans being able to actually understand what an AI agent did and why. That’s much more possible when the system is built on open components, such as open agent scaffolding, open rule engines, and auditable decision logs and traces, than when it’s a black box. The “human in the loop” is only meaningful if the human can see into the loop.
Companies don’t have to build these capabilities entirely from scratch. There’s a rich open-source ecosystem of security tooling, including vulnerability scanners, intrusion detection systems, log analyzers, and fuzzing frameworks, that AI agents can be integrated with.
For high-stakes organizations, starting from open, auditable foundations means security teams can actually inspect how their monitoring works, rather than trusting a single vendor’s claims. This is particularly important where sensitive data and processes are involved, and where sensitive material should generally not be flowing through external AI providers. Open systems can be rigorously analyzed by in-house security professionals, fine-tuned on an organizations’s own secure data, modified to produce organization-specific oversight mechanisms, and run entirely within an organization’s own infrastructure, keeping everything behind appropriate firewalls.
Attackers will develop models that take advantage of vulnerabilities. A significant part of the answer is leaning into transparent practices: open security reviews, published threat models, shared vulnerability databases, and open tooling that any team can adopt. The alternative of each organization trying to secure itself in isolation with proprietary tools doesn’t scale against attackers who are coordinating and sharing techniques in their own communities.
The future of AI cybersecurity will be shaped less by any single model and more by the ecosystems that surround them. Openness provides defenders with the visibility, the control, the community, and the shared infrastructure to stay ahead.
此内容由惯性聚合(RSS阅读器)自动聚合整理，仅供阅读参考。 原文来自 — 版权归原作者所有。