By Chris Green, OSCP, CISSP, CISA, CRISC, QSA/PCIP, CMMC RP, Structured GRC Director —

On April 13, 2026, Anthropic announced Project Glasswing, a cross-industry initiative involving AWS, Microsoft, Google, Cisco, CrowdStrike, Palo Alto Networks, JPMorganChase, Apple, Broadcom, NVIDIA, and the Linux Foundation.

At the center of the initiative is Claude Mythos Preview, an unreleased frontier AI model that has demonstrated a capability to find and exploit software vulnerabilities at a level matching or exceeding the world’s most skilled human security researchers.

The announcement is significant not because it introduces a new category of threat, but because it confirms that a threshold has been crossed, and the security community’s assumptions about attacker timelines, patch windows, and vulnerability discovery rates need to be revisited in light of it.

This isn’t a future risk. Mythos Preview has already autonomously discovered thousands of zero-day vulnerabilities across every major operating system and every major web browser, including flaws that survived decades of human review and millions of automated tests without being detected. Among the disclosed examples:

• A 27-year-old vulnerability in OpenBSD that allowed an attacker to remotely crash any machine simply by connecting to it,
• A 16-year-old flaw in FFmpeg, buried in a line of code that automated testing tools had touched millions of times without catching the problem, and
• Several chained vulnerabilities in the Linux kernel that, when combined, allowed escalation from ordinary user access to complete control of the machine.

All of these have been reported and patched, but the capability that found them is now established, documented, and will continue to develop.

The key takeaway is this: the window between a vulnerability being discovered and a working exploit being available has collapsed. What once took months now happens in minutes. The old calculus for patch timing, risk acceptance, and vulnerability management no longer holds.

What Mythos Preview Actually Means for Your Organization

The same capability Anthropic is deploying defensively will, over time, become accessible to actors with less scrupulous intentions. The advantage defenders currently hold is narrow and time-limited, and organizations that respond to this announcement as they would to a routine security advisory will find themselves behind. Three underlying shifts are driving this, and they affect every organization regardless of size, sector, or current compliance posture.

One: Vulnerability discovery is being automated at scale in a way that fundamentally changes the scope of your attack surface. Flaws that survived years of manual review and automated scanning are now being found in hours by models that can reason about code in ways no previous tool could. Your systems contain vulnerabilities that existing tooling has never surfaced, and that reality is now more immediately consequential than it was six months ago.

Two: Exploit development is accelerating alongside discovery, to the point where they are increasingly part of the same automated workflow. Finding a flaw and building a working exploit used to be sequential activities separated by time, skill, and effort. That separation is narrowing, and it changes the math around how long organizations have to respond once a vulnerability is identified, whether by defenders or by adversaries.

Three: Open-source software, which underpins most enterprise environments, has historically received less security scrutiny than commercial products, in part because open-source maintainers rarely have the resources that dedicated security teams at large vendors do. AI-assisted scanning is now exposing that gap rapidly, and the organizations most reliant on open-source components without a clear picture of what they are running are the most exposed.

What to Do Now

The following action items are organized by urgency, but they are not meant to be read as a checklist to satisfy and set aside. They are a starting framework for a conversation about whether your current security posture is calibrated for the environment as it now exists, rather than the one that existed a year ago. All of them are grounded in controls and frameworks your organization may already be working within, including NIST CSF v2.0, NIST SP 800-53, and applicable sector-specific standards.

Immediate: Audit your patch and vulnerability management process. The mean time to patch for critical and high-severity vulnerabilities must be compressed, and the process by which you prioritize and schedule patches needs to reflect that. Review your current SLAs against the detection and response controls in NIST CSF v2.0 (DE.CM and RS.MI in particular) and honestly assess whether the timelines you have accepted are still defensible given how quickly exploitation can now follow discovery.

Immediate: Confirm the completeness of your software inventory. AI-assisted scanning surfaces vulnerabilities in components that many organizations do not know they are running, because their software inventories are incomplete or out of date. If you don’t have a current software bill of materials (SBOM) for your critical systems, that is the place to start, and it is worth noting that several federal and sector-specific frameworks already require one. This isn’t a new obligation, but it’s become more urgent.

Near-term: Assess your open-source and third-party dependency exposure. Most enterprise environments run significant open-source components, often several layers deep in their software stacks. Catalog what you depend on, establish a monitoring posture for CVE disclosures against those components, and make sure the teams responsible for patch decisions are aware of the exposure. This is particularly important for organizations that use managed services or third-party platforms, where the dependency chain may not be immediately visible.

Near-term: Review your logging, detection, and alerting coverage. Faster exploit timelines demand faster detection, and the gap between when an exploitation attempt begins and when your team knows about it needs to be as small as possible. Evaluate whether your SIEM or monitoring stack can realistically surface exploitation attempts within minutes rather than hours and identify any coverage gaps in the systems and environments that represent the highest risk.

Near-term: Revisit your incident response plan and tabletop schedule. If your IR plan was last exercised more than twelve months ago, it predates the current threat environment in meaningful ways. Schedule a tabletop exercise and make sure the scenario includes an AI-augmented attacker, because the assumptions about attacker speed and sophistication that underlie most legacy IR plans are no longer current.

Strategic: Begin aligning secure development practices to the new baseline. For organizations that develop software internally, NIST SP 800-218 (the Secure Software Development Framework) and broader secure-by-design principles have become more urgent. Code review processes, static analysis tooling, and developer training should all be evaluated against a threat model that includes automated, AI-assisted vulnerability discovery. The cost of remediating a vulnerability after deployment is substantially higher than catching it during development.

Strategic: Engage with your GRC or vCISO advisor on framework alignment. NIST CSF v2.0, SP 800-53 Rev. 5, and sector-specific frameworks all include controls that are directly applicable to this shift in the threat landscape. A structured gap assessment against those controls is a logical next step for most organizations, and it provides a documented basis for the prioritization decisions you will need to make.

Sector-Specific Considerations

Defense Contractors and CMMC-Regulated Organizations

Organizations operating under CMMC requirements, or working toward them, face a compounded risk in this environment. AI-assisted vulnerability discovery applied to systems that handle Controlled Unclassified Information (CUI) could accelerate the exposure of data with direct national security implications. The control families in NIST SP 800-171r3 most directly relevant here are vulnerability management (3.11.x) and incident response (3.6.x), and both should be reviewed with the understanding that the attacker timeline assumptions embedded in those controls were written for a different environment than the one that now exists. Configuration management (3.4.x) is also worth revisiting, particularly for organizations with complex or layered environments where unauthorized changes may be difficult to detect quickly.

Financial Services and PCI DSS Environments

For organizations subject to PCI DSS v4.0.1, Requirements 6 and 11 are the most immediately relevant, covering secure system development and testing respectively. The shift in exploit timelines raises the bar on how quickly patches must be applied within the cardholder data environment and how rigorously the scope of that environment is defined and maintained. Annual penetration testing remains a required control under most frameworks, but it is increasingly insufficient as a standalone practice, and organizations should consider whether more frequent targeted testing of high-risk components is warranted. For card-issuing organizations in particular, the combination of CDE boundary clarity and continuous monitoring is more important now than it was at the time PCI DSS v4.0.1 was published.

Healthcare and HIPAA-Regulated Organizations

Healthcare organizations carry a distinctive exposure in this environment, and it operates on two levels. The first is operational: healthcare systems are high-value targets because of the sensitivity of the data they hold and the operational consequences of disruption, and AI-assisted exploitation lowers the bar for adversaries who previously lacked the technical sophistication to successfully attack well-defended hospital networks. The second is regulatory. The HIPAA Security Rule requires covered entities and business associates to conduct regular risk analyses (45 CFR 164.308(a)(1)) and to implement risk management processes that respond to identified risks in a timely way. A shift of this magnitude in the threat landscape constitutes a material change to the risk environment, and organizations that have not updated their risk analysis documentation in response to it are carrying regulatory exposure in addition to technical exposure. The Department of Health and Human Services Office for Civil Rights has consistently emphasized that risk analysis must be current and accurate, and a risk analysis that does not account for AI-augmented threat actors is neither. Healthcare organizations should also pay particular attention to legacy systems and medical devices, which often carry long patch cycles and may be running open-source or unmaintained software components that are now more exposed than they were previously understood to be.

Government Agencies and Critical Infrastructure

Anthropic’s announcement explicitly names critical infrastructure as a priority focus for the defensive use of Mythos Preview, which is also a signal about where adversary attention will be directed as similar capabilities proliferate. State and municipal systems, utilities, transportation networks, and other critical infrastructure operators should review their continuous monitoring posture and the currency of their risk assessments, with particular attention to legacy systems that may not have been updated in years. Applicable frameworks including CJIS, FedRAMP, FISMA, and sector-specific standards all include vulnerability management and continuous monitoring requirements that should be revisited in light of compressed exploit timelines. For government clients that haven’t yet conducted a tabletop exercise incorporating AI-augmented threat scenarios, doing so in the next ninety days is a reasonable and defensible step.

Privacy Law and General Regulatory Obligations

Organizations subject to general privacy laws, including GDPR, CCPA/CPRA, and the growing body of state-level privacy legislation in the United States, have obligations that extend beyond the sector-specific frameworks discussed above. Most of these laws require that organizations implement reasonable security measures appropriate to the risk, and the standard for what is “reasonable” evolves as the threat landscape does. The emergence of AI-assisted vulnerability exploitation is precisely the kind of material development that regulators and plaintiff attorneys will point to when assessing whether an organization’s security practices were adequate at the time of a breach. Documenting that your organization was aware of this development and took deliberate steps in response to it is therefore not just a security practice but a legal risk management practice. Organizations handling large volumes of personal data, sensitive categories of data such as health information, financial records, or biometric identifiers, or data belonging to residents of high-scrutiny jurisdictions should treat this announcement as a prompt to review their data security practices, update their records of processing activities where relevant, and confirm that their vendor and processor agreements appropriately address security obligations given the current environment.

A Final Note on Mythos Preview Perspective

Project Glasswing is a defensive initiative, and the organizations involved are working to use this capability to find and fix vulnerabilities before adversaries can exploit them.

The announcement is not a cause for alarm in the sense of suggesting that attacks are imminent or that existing defenses have suddenly failed. What it represents, rather, is a clear signal that the bar has moved, and that organizations which have deferred security fundamentals, whether patch management discipline, software inventory completeness, logging coverage, or incident response readiness, are now more exposed than they were, and the window for addressing that exposure proactively is a narrow one.

The right response is deliberate, prioritized action on fundamentals that are well understood and grounded in frameworks most organizations are already working within. Structured’s GRC Advisory and vCISO practice is available to assist with prioritization, gap assessments, framework alignment, and advisory support as you think through what this means for your specific environment.

Does your organization need help getting back to its cybersecurity and compliance basics as a result of what has been revealed by Claude Mythos Preview? Contact your Structured account manager or email info@structured.com today.

About the Author

Chris Green, CISSP, CISA, OSCP, CRISC, QSA/PCIP, CMMC RP, is an experienced cybersecurity professional with over a decade in the field, focusing on offensive security, risk assessment, and compliance. As Structured’s GRC Director, Chris leads a team of elite compliance and penetration testing professionals in identifying and mitigating security vulnerabilities across complex environments. His expertise spans network and web application penetration testing, phishing and social engineering engagements, and PCI DSS compliance assessments as a Qualified Security Assessor (QSA).

Chris holds several certifications, including the OffSec Offensive Security Certified Professional (OSCP), ISC2 Certified Information Systems Security Professional (CISSP), ISACA Certified Information Systems Auditor (CISA), and Certified in Risk and Information Systems Control (CRISC), and more, demonstrating his ability to perform thorough penetration tests, audit a variety of environments against compliance standards, and manage and assess IT risks. In addition to hands-on technical skills, he provides virtual CISO services, guiding organizations in aligning security strategies with business objectives.