Fabraix has launched Nyx, an autonomous red-teaming agent that can run adversarial attacks on any AI agent using a library of over 10,000 jailbreaks to uncover security vulnerabilities. The tool supports chat, voice, browser, and coding agents, and reports a 78% attack success rate on AgentHarm, with exploitable findings in agents at dozens of Fortune 500 companies.
SynthePulse Insight · AI deep reading
Version 1 · 1 source
Fabraix launches Nyx, an autonomous red teaming agent that claims a 78% attack success rate on the AgentHarm benchmark and has found exploitable issues in dozens of Fortune 500 companies. This marks a key step in the shift from manual to automated AI security testing.
According to a post on X by TestingCatalog, Fabraix has launched an autonomous red teaming agent called Nyx. Users can point it at any AI agent and run adversarial attacks from a library of over 10,000 jailbreaks to uncover security vulnerabilities.
Nyx covers chat, voice, browser, and coding agents, and requires only an endpoint, URL, or phone number as input. This lowers the barrier to entry, making security testing more accessible.
Fabraix reports that Nyx achieves a 78% attack success rate on the AgentHarm benchmark and has found exploitable vulnerabilities in dozens of Fortune 500 companies. These figures come from Fabraix's official statements and have not been independently verified.
Notably, these findings suggest Nyx's effectiveness in real-world environments, but specific vulnerability details and testing methodologies were not disclosed in the source.
Each of Nyx's findings outputs the attack steps, enabling agents to identify vulnerabilities. This explainability helps security teams understand attack paths and remediate issues.
Additionally, the same tests can be replayed after release to check if vulnerabilities have been fixed. This provides a mechanism for continuous security validation.
Following the release, one user commented, "Red teaming agents red teaming agents, we've built an entire economy where AI checks AI's homework," reflecting attention on the trend of AI self-security testing.
Nyx's release may signal a shift from human-led to automated AI security testing, but more independent validation is needed to confirm its claimed effectiveness.
This report is based on a post by TestingCatalog on X, a third-party news source, not an official Fabraix announcement. All performance data about Nyx (such as the 78% success rate) and commercial validation (Fortune 500 findings) come from Fabraix's reports and have not been independently verified; they should be treated as source claims rather than confirmed facts.
Nyx represents a new direction in automated AI security testing, but its claimed high success rate requires independent verification. For security teams, replayability and explainability are its potential value.
Primary report
Primary source