Adversa AI
Runtime security for AI coding agents, plus AI red teaming
Overview
Adversa AI (Adversa AI Ltd, Tel Aviv) sells a Coding Agent Security Platform, in early access, that watches and blocks dangerous actions by Claude Code, Copilot, Cursor and Codex. It also offers scoped red-teaming of models, agents and MCP servers, publishes attack research and holds a US patent filed in 2019.
Pricing
Pricing shown for reference only. These figures reflect RECATOOLS research as of 12 Oct 2026 and may be out of date or incomplete. This is not financial or purchasing advice — always confirm the current price on the provider’s official website before making any decision.
- Coding Agent Security Platform pilots, observe-first, run with the founding team
- Scoped red-teaming engagements for models, agents and MCP servers
Use cases
What you can produce with Adversa AI
- An observe-mode log of every model call, tool call and file action your coding agents take
- Blocking of risky action chains, such as a secret read followed by a public push, before they complete
- Agent policies written in plain English and compiled into enforceable rules
- Scoped red-teaming of your LLM app, agents or MCP servers, with findings ranked by business impact
- Audit evidence mapped to the EU AI Act, NIST AI RMF, Singapore's Model AI Governance Framework and MAS FEAT
- OpenClaw agents hardened with the open-source SecureClaw suite
- AIRQ security ratings of AI agents, with a published methodology
Adversa AI has moved from red-teaming language models to a runtime control layer for AI coding agents, which links model calls, tool calls and endpoint actions into chains and blocks dangerous ones; red teaming continues as a scoped service. Its published attack research, including 2026 Copilot CLI and Cursor findings, is real and current.
The new platform is in early access with founder-run pilots, nothing is priced publicly, and customer and award claims are self-reported; Gartner Peer Insights shows 4.2 from 13 ratings. Its compliance mapping includes Singapore's Model AI Governance Framework and MAS FEAT, which matters to regulated ASEAN teams.
What we found
Adversa AI is a Tel Aviv security company, registered as Adversa AI Ltd, that has changed what it leads with. The site now opens on a Coding Agent Security Platform: a runtime layer that watches what Claude Code, Copilot, Cursor and Codex actually do, links model calls, tool calls and endpoint actions into chains, and blocks a chain before it completes. That platform is in early access, with pilots "scoped, observe-first, and run with the founding team". The AI red teaming the company built its name on is still sold, now framed as a scoped service for models, agents and MCP servers.
The research record is long. Adversa's US patent on protecting AI applications against attacks was filed in its own name in September 2019, and it has published offensive work for years, most recently a GitHub Copilot CLI secrets-exfiltration attack in October 2026 and a Cursor deeplink flaw in July. In 2026 it released SecureClaw, an open-source security suite for OpenClaw agents, and AIRQ, a rating of 100 enterprise AI agents built by Adversa itself rather than an outside body. The site names Daniel Rubinstein as co-founder and chief executive and Alex Polyakov as co-founder and chief technology officer; Polyakov co-leads the Coalition for Secure AI's agentic security workstream.
Outside evidence is thin. Gartner Peer Insights showed 4.2 from 13 ratings on 12 October 2026, twelve of them four or five stars, and lists two to ten employees. The awards and customer claims on its site, such as "Fortune 500 | Big 4 | Decacorns", are Adversa's own statements.
There is an ASEAN angle: the coding-agent platform maps its evidence to Singapore's Model AI Governance Framework and MAS FEAT, which matters to regulated teams. There is no price list or self-serve tier; everything starts with a demo request.
It suits security teams already running coding agents or production LLM apps who want testing and runtime controls from researchers who publish their attacks.
Compiled by RECATOOLS from public sources: vendor documentation and pricing, code repositories and published reporting.
About this listing
This entry was compiled from publicly available data including Adversa AI's official website, press releases, documentation, and reputable third-party publications. RECATOOLS is not affiliated with Adversa AI unless explicitly stated.
Third-party AI tools update their pricing, features, availability, and policies frequently. Information here may be outdated by the time you read this — we make reasonable efforts to keep listings current, but cannot guarantee absolute accuracy.
For the latest details, please refer to Adversa AI directly →
Spotted something out of date? Suggest an update →
Alternatives to Adversa AI
More in Security & Safety
