Early access · CounterICE Secure
Defense begins
with offense.
“The best defense in fighting is an aggressive defense.”Jack Dempsey · Championship Fighting, 1950
Agentic Security Validation Swarm
Coordinated agents. Independent verification. Researcher judgment.
Multi-agent. Multi-model. One coordinated investigation.
Attack in parallel.
Learn as a swarm.
Secure coordinates agents around a shared objective. A planner breaks the target into investigations; workers explore different attack paths, publish evidence, and give the next agent a stronger starting point. Researchers define the scope and help turn the results into decisions.
Why proactive security
Find the path in.
Prove what is at risk.
Every release changes the attack surface. Secure maps the exposure, tests attack hypotheses, and challenges the evidence. Repeat the investigation when your code changes—not months after the last assessment.
Research & open work
Open research.
With thanks.
Our write-up on the workbench, and the open research on agent harnesses and autonomous penetration testing that shapes how we think about Secure. Results are as each paper reports them.
Our write-up on the workbench behind Secure: scoped Web2 investigation, EVM audits with Diff / Replay, the Solana preview, and what it takes for a finding to count as confirmed.
AgentFlow. Searches the whole harness — agent roles, prompts, tools, communication and coordination — and reads runtime signals from the target to decide which part to rewrite. Reports 84.3% on TerminalBench-2 and ten previously unknown vulnerabilities in Google Chrome, two of them Critical sandbox escapes.
Multi-agent web penetration testing with persistent memory, vulnerability-specific pipelines and browser-backed verification, in place of unrestricted exploration. Strongest on injection: reports 87% XSS success on the XBOW benchmark.
Shared memory, retrieval and a two-phase reflection step that validates the agents' results. Reports 86% overall success on the XBOW benchmark.
Ten frontier models from seven providers, run through a multi-agent framework on all 200 NYU CTF Bench challenges. The environment's tooling and the choice of model moved results more than prompting variations did.
The other side of agent security: the API routers between an agent and its model. Measures payload injection and secret exfiltration by malicious routers, and evaluates client-side defenses.
-[SECURE]- · Enterprise partnership
Keep security moving.
On your terms.
Plan repeat security reviews as your sites and projects change. Security scans surface signals; agentic investigation adds context for your team to review.
Configured for your business.
Start with a one-time enterprise setup fee. We configure Secure around your environment and agreed scope.
Focused on your assets.
Secure is restricted to your own websites and projects, keeping each scan within the scope your enterprise has agreed.
Run on your schedule.
Choose when repeat investigations and scans run. Discuss deployment and integration around your release process.