Zero-Day Vulnerabilities Leaked: 15 Exploits Published by Anonymous Researcher

An anonymous researcher has published working exploit code for 15 zero-day vulnerabilities—including two already under active attack—without prior disclosure to affected vendors, raising urgent risks for developers, enterprises, and open-source maintainers.

The dump, posted in a now-removed GitHub repository called exploitarium, includes proof-of-concept code for flaws in widely used software like libssh2 (CVE-2026-55200) and Gitea (CVE-2026-20896). Both vulnerabilities are rated critical and have seen exploitation in the wild, according to Federal Signal analyst Ethan Andrews, who built detection rules for the disclosed flaws.

Libssh2, a client-side SSH library used in tools like PuTTY and WinSCP, contains a pre-authentication heap corruption bug that allows remote code execution. A patch exists in the project’s development branch but has not yet been released. Gitea’s vulnerability, meanwhile, enables authentication bypass in self-hosted Docker deployments, letting attackers impersonate any user and take full control of the Git server. The fix was released in Gitea 1.26.3 on June 28.

The researcher, who uses the handle bikini, claims none of the 15 vulnerabilities were reported to vendors before publication. Their approach mirrors that of Nightmare Eclipse, another anonymous bug hunter who has leaked Microsoft zero-days in recent months. Unlike Nightmare Eclipse, however, bikini’s targets span multiple industries—including Splunk, RustDesk, 7-Zip, VLC, AnyDesk, OpenVPN, c-ares, and Floci—without apparent vendetta against any single company.

Andrews and others speculate bikini may have used AI-assisted fuzzing, specifically GPT-5.5 Codex, to automate vulnerability discovery. While some disclosed flaws have been dismissed as low-impact "AI noise," Andrews’ analysis confirms two high-risk vulnerabilities with confirmed exploitation:

• libssh2’s heap write flaw (CVE-2026-55200)
• Gitea’s Docker auth bypass (CVE-2026-20896)

The exploitarium repository, though removed by GitHub, persists in archives and dark-web forums. Attackers are likely scanning for vulnerable instances using AI tools, reducing the time needed to weaponize the disclosed flaws. Federal Signal’s 44 KQL detection rules (with language translations for non-KQL stacks) aim to help organizations mitigate risks, but the lack of prior vendor coordination complicates response efforts.

Why is responsible disclosure being bypassed—and what are the consequences?

Responsible disclosure, the practice of reporting vulnerabilities to vendors before public release, is widely considered the gold standard in cybersecurity. It allows developers to patch flaws before attackers exploit them. However, bikini’s approach undermines this model by publishing exploits without warning, forcing organizations to scramble for fixes.

The immediate impact includes:

• Unpatched systems at risk: Organizations using affected software (e.g., libssh2, Gitea) must manually apply fixes or wait for vendor releases.
• Increased attack surface: Exploit code in the wild accelerates exploitation, as seen with CVE-2026-55200 and CVE-2026-20896.
• Erosion of trust in open-source security: Maintainers of projects like libssh2 and Gitea now face pressure to accelerate patch cycles, even for critical flaws.

Federal Signal’s Andrews notes that while some disclosed vulnerabilities may be "noise," the two confirmed exploits demonstrate real-world danger. "The most technically significant findings have been independently verified as high-risk with active exploitation observed," he wrote, adding that bikini’s PoCs eliminate the need for attackers to develop their own exploits.

How do bikini’s methods compare to Nightmare Eclipse—and what does this mean for AI in cybersecurity?

Bikini’s actions parallel those of Nightmare Eclipse, who has leaked Microsoft zero-days in recent months. However, key differences emerge:

While Nightmare Eclipse’s leaks have drawn scrutiny for targeting a single vendor, bikini’s broader approach raises questions about AI’s role in vulnerability discovery. Andrews’ analysis suggests bikini may have used automated fuzzing powered by AI, a trend that could reshape how flaws are found—and exploited.

The AI-induced "vulnpocalypse" warning, while dramatic, reflects growing concerns. If AI tools can rapidly identify and weaponize vulnerabilities without human oversight, the cybersecurity landscape may shift from responsible disclosure to race-to-exploit dynamics.

What should organizations do now?

1. Patch immediately: Apply fixes for libssh2 (awaiting official release) and Gitea 1.26.3+.
2. Deploy detection rules: Federal Signal’s 44 KQL rules (available [here]) can help identify exploitation attempts.
3. Monitor for new leaks: Assume bikini’s exploits are already circulating in underground forums.
4. Review open-source dependencies: Audit third-party libraries for unpatched flaws.
5. Engage with maintainers: Open-source projects like libssh2 rely on community contributions to patch critical bugs.

The exploitarium dump serves as a stark reminder that zero-days can emerge from unexpected sources—whether through traditional bug hunting, AI-assisted fuzzing, or deliberate disclosure bypass. For enterprises and developers, the lesson is clear: assume compromise and act fast.

Sources and further reading

• Federal Signal analysis (Ethan Andrews’ detection rules): Federal Signal GitHub
• Gitea security advisory: Gitea 1.26.3 release notes
• Libssh2 patch status: Libssh2 GitHub
• Original reporting: The Register – "Anonymous researcher drops 0-day ‘exploitarium’ repo"