In April 2026, Anthropic launched Project Glasswing — a cybersecurity initiative that fundamentally changes what organizations should expect from their security programs. At the center is Claude Mythos Preview, an unreleased frontier model that has already identified thousands of previously unknown high-severity vulnerabilities across every major operating system, every major web browser, and critical infrastructure software used by billions of people.
Some of these vulnerabilities are over two decades old. A 27-year-old TCP SACK flaw in OpenBSD — an operating system specifically built for security, audited continuously by experienced researchers. A 16-year-old bug in FFmpeg's H.264 codec, missed despite more than five million automated fuzzing iterations. Multiple Linux kernel privilege escalation chains that survived decades of expert review.
Mythos found them. Human teams did not.
This is not a research paper or a proof of concept. Anthropic committed $100 million in model usage credits plus $4 million in direct grants to open-source security organizations, and assembled twelve founding partners: AWS, Apple, Anthropic, Broadcom, Cisco, CrowdStrike, Google, JPMorganChase, the Linux Foundation, Microsoft, NVIDIA, and Palo Alto Networks. Over forty additional organizations that build or maintain critical infrastructure also have access. Anthropic has stated that over 99% of the vulnerabilities Mythos has identified remain unpatched, and a public disclosure cycle is expected to land in early July 2026.
The clock is real. The window between "these tools exist" and "patches must be deployed" is measured in months — not years.
The Capability Gap Is Real
On CyberGym — a benchmark for real-world vulnerability reproduction — Mythos Preview scored 83.1%, compared to 66.6% for Claude Opus 4.6, the previous best public model. But benchmarks only tell part of the story. What matters is what Mythos does that human teams cannot: it develops sophisticated exploits autonomously, without human direction, across codebases that experts have reviewed for years.
As Cisco's SVP and Chief Security & Trust Officer Anthony Grieco put it: "AI capabilities have crossed a threshold that fundamentally changes the urgency required to protect critical infrastructure from cyber threats, and there is no going back."
This is not a warning about the future. This is a statement about today.
The Risk Calculus Just Changed for Every Enterprise
Whether you operate SAP, Oracle, custom applications, or a mix, Glasswing changes your risk posture in three ways:
First, your vulnerability backlog just got exposed. Tools built on Mythos-class models will find flaws that current penetration testing and SAST/DAST scanning tools cannot detect. Decades-old bugs sit in custom code, in integration interfaces, in authorization configurations that have been "good enough" for years. The question is not whether these vulnerabilities exist in your landscape. They almost certainly do. The question is who finds them first.
Second, your security vendors are already integrating these capabilities. CrowdStrike, Palo Alto Networks, and Cisco are Glasswing founding partners. Their products will incorporate Mythos-class AI within months. Your security stack is about to change whether you plan for it or not. The organizations that prepare now — with governance frameworks, testing protocols, and clear policies — will adopt these tools confidently. The organizations that react will scramble.
Third, governance becomes the differentiator. When AI can find vulnerabilities faster than human teams can triage them, the bottleneck shifts from detection to decision-making. If a Mythos-class tool surfaces 500 critical findings in your environment in a single assessment, do you have the framework to prioritize them? The change management process to patch without breaking production? The audit trail to demonstrate to regulators that your security decisions were sound?
The Governance Gap
This is where most organizations will struggle. Mythos is powerful, but power without governance is risk.
Consider the practical implications. An AI security tool runs against your enterprise environment and identifies 47 vulnerabilities across application extensions, API endpoints, and custom code. Twelve are rated critical. Three require immediate remediation. But the tool cannot tell you which ones will break your month-end close if you patch them Thursday. It cannot tell you which ones are acceptable risk given your specific business context. It cannot navigate the conversation between your security team, your basis or platform administrators, and your business process owners about what gets fixed in what order.
That conversation requires human judgment, organizational context, and governance frameworks that most enterprises have not built for AI-speed security findings.
CrowdStrike CTO Elia Zaitsev captures the operational reality precisely: "The window between a vulnerability being discovered and being exploited by an adversary has collapsed. What once took months now happens in minutes with AI." Closing that window is a governance challenge — not just a tooling one.
What This Means for Your SAP Landscape Specifically
If you run SAP, the picture is sharper still — because SAP itself is moving fast on Business AI, and the new attack surface is being built into your environment in real time.
At SAP Sapphire 2026 (May 11–13, Orlando), SAP is expected to push deeper on SAP Business AI, AI Foundation (the operating system for Joule and Joule Agents on SAP BTP), Business Data Cloud, Sovereign Cloud, and the AI Agent Hub in LeanIX. SAP CEO Christian Klein has already signaled that the company plans to "govern the agentic AI layer for our customers" — a clear acknowledgment that agent governance is itself becoming a first-class enterprise concern.
Every Joule extension, every custom agent built through Joule Studio, every API connection between Joule and your core ERP, every MCP gateway integration with non-SAP systems represents a surface that Mythos-class tools can — and will — probe. Custom ABAP code, RFC interfaces, BTP applications, Joule agent configurations. The organizations that govern these surfaces proactively will lead. The organizations that deploy first and govern later will pay the price in breaches, compliance failures, and regulatory scrutiny.
The Sapphire 2026 announcements are not just product news. They are an inflection point in the SAP customer's threat surface — and they arrive in the same quarter as the Glasswing public disclosure cycle.
The Restricted Model Question
Anthropic made a deliberate decision to restrict Mythos. It is not publicly available. The eventual goal is to enable safe deployment at scale, but today, only vetted organizations working on defensive security have access. Mythos Preview itself is priced at $25 per million input tokens and $125 per million output tokens for Glasswing participants — restricted in price as well as access. This restriction is itself a governance lesson: even the creators of the most powerful security AI recognize that capability without controls is dangerous.
For enterprises, the implication is clear. When your security vendors ship products built on Mythos-class models — and they will — you need to be ready with:
- Policies for AI-assisted vulnerability detection and remediation
- Incident response procedures calibrated for AI-speed discovery
- Audit trails that satisfy SOX, GDPR, and industry-specific requirements
- Clear ownership of AI security tool governance (vCISO, CISO, or dedicated AI governance role)
- Testing protocols for AI security tools themselves — who evaluates the evaluator?
What You Should Do Now
The window between "these tools exist" and "these tools are mandatory" is closing fast. Three steps every enterprise security leader should take this quarter:
1. Audit your current security stack. Identify which of your vendors are Glasswing partners or are building on frontier AI models. Understand how AI-powered capabilities will enter your environment — through your endpoint protection, your SIEM, your application security testing, or your cloud security posture management.
2. Assess your governance readiness. Run an honest gap analysis. Do you have policies for AI-assisted security findings? Change management processes that can handle AI-speed vulnerability discovery? Board-level reporting for AI security tool adoption? If the answer is "not yet" to any of these, you have work to do before the tools arrive — and before the July 2026 Glasswing disclosure wave hits your patch pipeline.
3. Engage an advisor. This is not a technology problem you can solve with another product purchase. It is a governance challenge that requires expertise in both cybersecurity and AI risk management. The right advisor helps you build the frameworks, train your teams, and establish the processes that turn AI security power into organizational advantage — not organizational chaos.
UX4Tech: Your Governance and Assessment Partner
UX4Tech exists to close the governance gap between AI capability and organizational readiness. We are not building competing security AI. We are the governance and advisory layer that helps enterprises adopt tools like Mythos responsibly — and prepare their environments to withstand whatever comes next.
How a UX4Tech Security Assessment Prepares You
Our security assessments are built around frameworks your auditors already trust and your SAP teams already speak. We map every finding and recommendation to:
- The SAP Secure Operations Map — SAP's own five-layer framework spanning Environment, System, Application, Process, and Organization. Every customer-side recommendation SAP publishes — security hardening, secure SAP code, security monitoring and forensic, identity and authorization, custom code security, compliance, and security governance — lives on this map.
- The NIST Cybersecurity Framework (CSF) — the global standard SAP itself aligns to, with its Identify, Protect, Detect, Respond, and Recover functions.
- AIUC-1, NIST AI RMF, and the CSA Agentic Trust Framework — the emerging standards for evaluating AI agents themselves, including the new Q2-2026 controls covering MCP security, agent permissions, and third-party agent risk.
The result: when our assessment surfaces a finding, you can trace it directly to a control SAP recommends to its own customers — and to a function your CISO already reports against. Your remediation plan reads as an extension of SAP's published methodology, not a parallel framework you have to defend separately to auditors, your board, or your basis team.
Where We Focus
Virtual CISO (vCISO) advisory — We embed with your security leadership to build AI-ready governance frameworks. Not templates. Working systems calibrated against your specific landscape, your regulatory environment, and your organizational culture.
SAP security assessment — Comprehensive evaluation of your SAP environment against the new threat landscape, organized along the Secure Operations Map. Authorization and identity analysis, RFC and gateway security, custom ABAP code review, Joule and Joule Agent governance, BTP and AI Foundation platform security, MCP gateway and AI Agent Hub configuration, and integration security across SAP and non-SAP systems — all calibrated for an environment where AI-powered attack tools are no longer hypothetical.
AI agent trust evaluation — As AI agents become embedded in enterprise operations, who evaluates whether those agents are trustworthy? We apply standards-based evaluation frameworks aligned with NIST AI RMF, the CSA Agentic Trust Framework, and AIUC-1 to verify that the AI tools you deploy meet enterprise security and governance requirements before they touch production data.
We are aligned with the Glasswing mission of securing critical software for the AI era. We help every organization — not just the ones with early Mythos access — prepare for what comes next.
In the AI era, audit-defensible governance is the new perimeter. The organizations that can document why a vulnerability was prioritized, deferred, or accepted — at AI speed — will lead. The rest will be explaining themselves to regulators.
Ready to assess your Glasswing readiness?
Schedule a 30-minute scoping call with our team.
Schedule Assessment →
