Researcher Details Release of Secret Zero-Day Exploits

A pseudonymous security researcher has released more than 30 proof-of-concept exploits for zero-day vulnerabilities across major open-source projects, including the Linux kernel, Libssh2, FFmpeg, Gogs, Gitea, Ghidra, 7-Zip, MyBB, PHP, OpenVPN, and VLC, without first notifying the maintainers. The dump, dubbed ‘Exploitarium’, was posted publicly on GitHub by an individual using the monikers ‘bikini’ and ‘ashdfrkl’ on Discord.

First published on June 27, the repository initially contained around 15 exploits, with the researcher adding new entries over the following days. In the repository, bikini claimed the entire fuzzing process was automated using AI, specifically OpenAI models and tools. Fuzzing, a common automated testing technique, involves feeding random, invalid, or unexpected data into software to trigger crashes, memory leaks, or security flaws.

The primary controversy surrounding this exploit dump stems from the apparent lack of coordinated vulnerability disclosure (CVD). CVD is the standard industry practice of privately alerting developers to a security flaw, granting them time to issue a patch before the vulnerability becomes public. On GitHub, bikini explicitly invited others to file CVEs themselves, framing the release as an effort to attract newcomers to the field.

Speaking to Infosecurity on Discord, the researcher confirmed they did not inform any project maintainers about the publication. While they have participated in CVD processes before, they chose not to this time. “I think it’s the best way for people to learn and become allured into the field. It’s a lot less interesting and informative if someone has to read a write up that’s not applicable by today’s security standards,” bikini said. “It also raises the barrier to entry making someone go back and install outdated software to test on.”

Some of the vulnerabilities have since been publicly disclosed and patched. One notable example is CVE-2026-55200, a severe pre-authentication remote code execution (RCE) vulnerability in libssh2, a widely used client-side C library for the SSH2 protocol. With a CVSS severity score of 9.2, exploitation involves sending specially crafted SSH packets with oversized packet_length values to manipulate heap memory, enabling remote code execution. Although bikini dropped the exploit on GitHub, the vulnerability was publicly disclosed by VulnCheck through formal channels, crediting researcher Tristan Madani (also known as @TristanInSec) for reporting it. A fix has been integrated into the libssh2 mainline development branch, though a formal release is still pending.

Ethan Andrews, a cybersecurity analyst and detection engineer at Federal Signal Corporation, told Infosecurity that CVE-2026-55200 has been “independently verified.” He described it as the “most severe” vulnerability from the dump and noted it is experiencing active exploitation.

Beyond CVE-2026-55200, bikini’s repository lists 12 issues that have now received CVE identifiers, covering memory corruption in FFmpeg, heap buffer overflows in libssh2, a 7-Zip Mark-of-the-Web bypass, container escape in Gitea, privilege escalation in MyBB, HTTP request smuggling in nghttp2, remote input injection in RustDesk, case-sensitivity bypass in Flowise, integer underflow in Nmap, use-after-free in the Ladybird Web Browser, and authentication bypass in NodeBB.

As new entries land in the repository, Andrews has built 44 Kusto Query Language (KQL) detection rules, released on Detections.ai and GitHub. KQL rules are used in security tools like Microsoft Sentinel to identify and respond to threats. Andrews also noted that some issues raised by bikini “have been community dismissed as low impact noise.”

Commenting on the dump-when-ready approach, Andrews said, “It shows a meaningfully different intent than a coordinated offensive toolkit release, but a risky decision at the same time, especially with no vendor coordination.” Patrick Garrity, a vulnerability researcher at VulnCheck, emphasized that his company “strongly encourages a coordinated approach.” He added, “We provide coordinated vulnerability disclosure as a free service and we issue CVEs when we observe vulnerabilities in the wild that don’t have one.”

In the repository, bikini included a warning against malicious use: “Do NOT, under any circumstances, use any material in this repository maliciously. This is good-faith, open-disclosure vulnerability research intended to get more people interested in exploring this area of cybersecurity. Cybercrime is cringe.” When asked if this disclaimer would deter malicious actors, bikini responded, “Of course not. The disclaimer might help, but at the end of the day, they have the free will to make their own choices.”

Nevertheless, bikini argued that public disclosure “just speeds up the patching process and will get these issues resolved quicker, limiting attackers who might already be aware of these things.” They added, “I just came to the understanding that open disclosure is better for everyone in 99% of circumstances.” Garrity predicted that “we are going to continue to see more of these type of drops.” The approach mirrors that of Nightmare Eclipse, a zero-day bug hunter who published Microsoft exploits in May 2026.

Regarding the AI methodology, bikini claimed on GitHub to have used an OpenAI model, initially attributing the work to GPT-5.5-3-Codex-Spark before revising it to GPT-5.3. “You do NOT need a SOTA model to help you identify these issues, I promise!” they wrote. “While being able to afford a better model is helpful, my data seems to show that it is only marginal when paired with decent human oversight and a good harness. None of the actual PoCs themselves were vibe-coded; I did, in fact, hand-type them.” Bikini told Infosecurity they “didn’t face any issues with AI safeguards,” but the real challenge is to “find bugs that interest people.” They plan to publish more details on their workflow in the future, adding, “I think it’s important to establish your own workflow first of what you’ve found to work best and implement a strict pathway for AI to automate this process for you.”

Infosecurity contacted maintainers of libssh2 and Ghidra but received no response at the time of publication.