the universal allowlist
for agents.
frontier models miss 57% of threats in the context your agents consume. brin pre-scans packages, skills, and web pages to detect malware, prompt injection, and supply chain attacks. free.
every context type
six context types, each with a dedicated threat model and scoring methodology. if your agent can reach it, brin scores it.
sub-10ms
pre-scanned results return in under 10ms. fast enough to sit in the critical path of every agent action — no queues, no cold starts.
one http call
no sdk, no auth, no signup. a single GET request returns a score, verdict, and threat data. add to any agent or pipeline.
##securing context, not agents
the default approach to agent security is guardrails — restricting what the agent can do. block certain tools, sandbox file access, limit network calls. it works, but it also cripples the agent. the more you constrain it, the less useful it becomes.
brin takes a different approach. let agents be as unconstrained as possible — and instead score every piece of external context they interact with. the risk was never the agent. it's the external context the agent trusts by default.
in brin-bench, claude opus missed 57% of the threats brin identified — including 100% of graph-based signals like dependency chains and publisher reputation. the model sees content. brin sees identity, behavior, and trust.
secure the context, not the agent. you get safety without sacrificing capability.
##what we score
six types of artifacts that agents consume autonomously — each with a distinct threat model and scoring methodology.
web pages
prompt injection, phishing, cloaking, hidden exfiltration
70% missed by models
packages
install-time attacks, credential harvesting, typosquatting
71% missed by models
repositories
agent config injection, malicious commits, compromised dependencies
89% missed by models
skills
description injection, output poisoning, instruction override
22% missed by models
pull requests
author risk, prompt injection, secret exposure, CI sabotage
benchmark coming soon
contributors
sleeper accounts, typosquat identities, anomalous activity
60% missed by models
##how it works
before your agent acts on any external context, make a single GET request. brin returns a score, verdict, and any detected threats. if brin is unreachable, the agent continues as normal — zero risk to your existing workflow.
##latest findings
real threats detected across the ecosystem.
every score brin produces is public and the tools are open-source. the api requires no auth, costs nothing, and handles 300 requests per minute per ip. if you prefer not to depend on the api, download the full dataset and host it on your own infrastructure.
the more widely trust signals are available, the harder it becomes for malicious context to spread. open data makes the entire ecosystem safer.
start scoring agent dependencies.
integrate brin in minutes — one GET request is all it takes. query the api, browse the registry, or download the full dataset.