The "Al Vulnerability Storm”: Building a "Mythos-ready" Security Program
Executive Summary
Mythos AI sharply increases attackers' offensive capabilities while only modestly aiding defenders. Attackers gain an asymmetric advantage and the defender burden grows disproportionately.
Near-term, security teams will be overwhelmed. Glasswing's disclosures are the first wave, and Mythos-level capabilities will proliferate, driving more frequent, complex, novel attacks.
Absorb the current spike
- Adjust risk calculations for rising patch volume and shorter time-to-patch
- Re-orient resources for more persistent, complex attacks
- Harden via segmentation, egress filtering, MFA, defense-in-depth
Be ready for the next waves
- Robust third-party and open-source dependency management
- Automated security assessments in dev; LLM agents find vulns first
- AI agents in the cyber workforce, to match attacker speed
- Re-evaluate tolerance for remediation downtime under shorter adversary timelines
- Update governance for faster vendor onboarding and AI defense deployment
Mythos & Glasswing: Why They Matter
To act on that asymmetry, start with what Mythos actually does and where Glasswing fits.
Mythos
Mythos has three key capabilities:
- Exploits without scaffolding: Mythos generated 181 working exploits on Firefox where its predecessor succeeded only twice under the same conditions.
- Complex, chained vulnerabilities: It identifies vulnerabilities composed of multiple primitives chained together into a single exploit path.
- "One-shot” capability: It accomplishes significantly more with a single prompt, requiring less elaborate configuration.
Glasswing
Glasswing is Anthropic's coordinated-disclosure program giving partners early Mythos access for patching. The advantage is bounded. The world's attack surface dwarfs any curated ecosystem, and comparable offensive capabilities will proliferate.
The Evolution of LLM-based Offensive Capabilities, 2025/6
Mythos didn't appear out of nowhere. AI-driven offensive capability has been compounding for the past year, and the cadence is accelerating:
10 Questions to Understand Your Security Program State and Influence
Against that trajectory, where does your program stand, and where can you act? If you can't answer one with a concrete example, that's the first finding.
Key Takeaways for the CISO
Whatever the answers, these moves put a program in Mythos-ready posture.
Use LLMs for vuln discovery and remediation
Defensive capabilities are mature. Start with agent-based code reviews; build toward VulnOps.
Update risk metrics
The shifting landscape has outdated old assessments. Refresh them and communicate the change to stakeholders.
Accelerate the team with coding agents
Agents speed up incident response, GRC, patch triage, red teaming, and data collection.
Prepare for more incidents
Run tabletops for simultaneous incidents. Automate remediation; verify segmentation, egress filtering, phishing-resistant MFA.
Refocus on the basics
Segmentation, patching, IAM raise attacker difficulty. Expand these efforts.
Prepare for burnout
Disclosure cadence will be taxing. Secure headcount and budget for reserve capacity and automation.
Build a Mythos-ready program
Mythos is one of many shifts. Bake its implications into security strategy.
Build collective defense
Attackers crowdsource tools as syndicates; coalitions beat teams. Engage ISACs, CERTs, and sector groups to share intel, coordinate response, and produce sector-specific guidance.
References
- Cloud Security Alliance, The "AI Vulnerability Storm": Building a "Mythos-ready" Security Program (original report).