About this project
AI agents find vulnerabilities, write exploits, run intrusions, patch code, and defend networks. They are also a new attack surface. The work describing all of this is scattered across lab system cards, government evaluations, threat reports, security advisories, and preprints, and much of it goes stale within months.
The Agentic Cyber Explorer brings that record into one place and keeps it honest about what is known. It is built for security researchers, practitioners, and policy analysts who need to understand the field quickly, stay current without reading every feed, judge how much weight a claim can bear, and find the questions worth studying next. New here? Start with how to read this.
Preview status
The explorer is in preview. Every record has been checked against its sources, but most have not yet had an editor’s final review: 0 of 219 records are marked human-reviewed. Each page shows its review state. Until review is complete, treat assistant-drafted records as well-sourced leads and follow the links before citing a detail.
Who maintains it
Fide AI is an independent research lab building more trustworthy AI and public tools for deciding when to trust it. The explorer is one of Fide’s public tools. The AI Alignment Explorer maps the broader AI safety field, and the Defense Evidence Lab is its companion for cybersecurity: where this explorer is the record of what has happened and what is known, the Lab is a place to practice judging what a piece of evidence can support. Pages here link to the Lab’s matching cases and research questions. The explorer supports Fide’s cybersecurity research on controlled autonomous response, verified repair, resilient defense teams, and whether security evaluations stay valid over time.
The responsible editor is Alex Chao (Founder, Fide AI), who approves what the explorer publishes. Updated at least every two weeks; every addition, revision, and correction is listed in the changelog.
How it is made, and the role of AI
Records are drafted with Anthropic’s Claude models from the sources they cite, then checked by software for schema, dates, cross-references, and duplicate sources. Additions enter the corpus only after an editor approves them. Findings, key questions, and openings are written the same way and carry the same review labels. Every record says whether a Fide editor has read it against its sources.
Anthropic, which makes the drafting models, is also one of the organizations the explorer covers most, as a developer, an evaluator, and a source of threat reports. To keep that from shaping the record, the same rules apply to every organization: claims are attributed to their source, an organization’s statements about its own systems are recorded as reported rather than measured, and superlatives such as “first” appear only when a source makes them. If you see coverage that favors or disfavors any organization, tell us.
Funding and independence
The explorer has no dedicated external funding. It is built and maintained by Fide AI as part of its cybersecurity research program, and no organization in the corpus pays for inclusion, placement, or review. The explorer follows Fide AI’s independence and conflict-of-interest policy, under which funders, clients, and partners may not direct findings or suppress unfavorable ones. Fide’s own research appears where it bears on the evidence and is labeled as Fide’s.
How to contribute
Suggest a source, report an error, or point us to prior work for an opening by writing to alex@fideai.org. Record pages include a prefilled correction link. The methodology explains what we include and how records are reviewed.
Reuse and citation
The explorer’s data is licensed CC BY 4.0; cite it as Fide AI, Agentic Cyber Explorer, with a link. Every record, finding, and key question has a citation and a JSON download. Follow new work through the RSS feed, a feed for any topic, or the Fide AI newsletter.
What it is not
It is not a threat feed, a vulnerability database, or a ranking of models or vendors. Nothing here is a Fide AI evaluation result unless it says so. No organization pays for inclusion or placement.