In April 2026, two events redefined cybersecurity: [Anthropic](https://anthropic.com) revealed that its model Claude Mythos had discovered thousands of zero-day vulnerabilities in all major operating systems and browsers, and [OpenAI](https://openai.com) responded with GPT-5.4-Cyber. The message is clear: AI has crossed a threshold where it outperforms the best human experts at vulnerability detection. Here is what every business needs to understand — and do.

What changed in 2026

Until 2025, AI in cybersecurity was mainly an assistance tool: log analysis, anomaly detection, malware classification. In 2026, we have shifted to autonomous offensive and defensive AI. Here are the three major changes:

1. AI finds flaws that humans cannot. Claude Mythos discovered a 27-year-old vulnerability in OpenBSD, a 16-year-old one in FFmpeg, and chained exploits in the Linux kernel — all autonomously.
2. Exploitation speed has skyrocketed. CrowdStrike estimates that the window between discovery and exploitation has shrunk from months to minutes with AI. Traditional patching approaches are no longer sufficient.
3. Defenders have the advantage (for now). The most powerful models are restricted: Mythos through Project Glasswing, GPT-5.4-Cyber through Trusted Access. But this lead is narrowing.

AI cyber tools available today

5 actions to take now

Whether you are an SMB or a large enterprise, here are the concrete actions to consider:

1. Modernize your security stack. Legacy tools do not detect AI-augmented attacks. Invest in solutions that themselves use AI for detection (CrowdStrike, Palo Alto Networks, etc.).
2. Accelerate your patching cycles. If the exploitation window is minutes, monthly patch cycles are no longer enough. Automate deployment of critical fixes.
3. Train your teams on AI. Your SOC analysts need to understand how attackers use AI and how to leverage it for defense.
4. Audit your open-source dependencies. AI models like Mythos find flaws in libraries everyone uses (FFmpeg, OpenSSL, etc.). Verify your software supply chain.
5. Apply for cyber program access. If you are eligible, apply to Anthropic's Cyber Verification Program or OpenAI's Trusted Access to get the most powerful tools.

Emerging threats: what is coming

Experts agree on several concerning trends for the next 12 to 18 months:

• Democratization of offensive capabilities. Open-source models will eventually reach cyber capability levels comparable to today's restricted models.
• Large-scale automated attacks. AI agents will be able to scan and exploit thousands of targets simultaneously, transforming the threat landscape.
• AI-augmented social engineering. Language models make personalized phishing accessible to any attacker, in any language.
• Sophisticated supply chain attacks. AI can identify subtle vulnerabilities in complex dependency chains.

The role of Project Glasswing

[Project Glasswing](/blog/claude-mythos-project-glasswing-cybersecurity) is the most ambitious initiative in the industry for using AI defensively. The consortium brings together AWS, Apple, Google, Microsoft, Nvidia, CrowdStrike, JPMorgan, and others to scan and secure the world's most critical software. Anthropic has committed to publishing a public report on the results within 90 days and producing practical recommendations on the evolution of security practices.

The areas covered will include: vulnerability disclosure processes, software update processes, open-source and supply chain security, secure development lifecycle, standards for regulated industries, and patch automation.

Frequently asked questions