Archestra reports that OpenAPPA blocked every scored attack across 1,320 evaluations on two security suites: Bench-Corp and AgentThreatBench. That is a striking result within the publisher’s stated tests—not proof that OpenAPPA is impossible to bypass or secure against every attack.
What does OpenAPPA’s 0% attack success rate mean?
On its 2026 evaluation page, OpenAPPA reports zero successful scored attacks in 1,320 guarded evaluations: 600 on Bench-Corp and 720 on AgentThreatBench. The runs include standard and adversarial prompts. In this context, 0% describes the observed result on those specified evaluations; it does not establish that future attacks, untested tasks, or different deployments will also fail.
As an Amazon Associate I earn from qualifying purchases.
The benchmarks cover distinct sets of tasks. OpenAPPA describes Bench-Corp as 20 multi-step workplace tasks, scored by checking task outcomes rather than using an LLM judge. AgentThreatBench is described as a 24-task suite. Each includes standard and adversarial testing, according to the project’s evaluation page. The results therefore apply to those suites, prompts, tools, models, policies, and scoring rules—not to every agent environment.
Can an agent still complete legitimate tasks?
Security is only useful if protection does not prevent routine work. OpenAPPA’s evaluation reports 89% task completion in its summary. Its detailed Bench-Corp table gives guarded OpenAPPA completion rates of 88.0%, 89.5%, and 90.0% across the three listed model rows.
#1 Best Overall
In that same detailed Bench-Corp table, the tested Microsoft FIDES configurations score between 37.0% and 44.5% for task completion, while their attack-success rates range from 28.0% to 34.5%, depending on model and configuration. Those figures describe the configurations and test conditions shown in OpenAPPA’s table; they are not a universal ranking of the products. The evaluation also says OpenAPPA enforced some requirements that were absent from the tested FIDES configurations, which is relevant when interpreting the comparison.
The page separately presents Claude Auto mode comparisons, with each task run once and no variance estimate. That comparison should not be read as evidence of statistical certainty. As with the main figures, model and configuration matter.
How does OpenAPPA enforce its policies?
Archestra describes OpenAPPA as open-source software that applies explicit information-flow policies to AI agents’ tool use. Its design tracks a trajectory label representing who may see information and how much it can be trusted, then checks each action against policy. The project says these checks are deterministic rather than asking another model to judge the agent’s intent.
Quick wins for a faster PC:
Repair Windows errors before they cause bigger problemsFix Now →Fix the driver behind crashes, sound loss and screen glitchesFind Drivers →When a policy blocks an action, OpenAPPA can provide a remedy plan. The project describes options such as sanitizing information, requesting scoped approval, or isolating work in a subagent. These are design claims from Archestra’s announcement and the product’s technical overview; they explain the intended enforcement model, not independent confirmation that every deployment is secure.
Does protection add token overhead?
OpenAPPA reports a 4.22% mean-token overhead over stock in a stated Tau Bench banking setup. Tau Bench is an ordinary task benchmark in this evaluation, not an attack benchmark. The figure therefore speaks to token use in that particular setup, not to a general production cost or evidence of attack resistance.
Independent reader supportYour contribution helps us test, update, and keep practical guides available for everyone.How strong is the evidence?
The 1,320-evaluation result is published by Archestra/OpenAPPA. No independent replication of this exact evaluation is established in the available reporting, so the finding should be treated as a vendor-published benchmark result. Benchmarks can reveal how a system performed under defined tests, but their coverage is bounded: a different model, policy, tool, prompt, or task can produce a different outcome.
Rank #4
Archestra’s September 28, 2026 announcement calls OpenAPPA “100% resistant” to certain data-exfiltration attacks. That is the company’s wording, not a conclusion supported beyond the finite benchmark scope. The more precise statement is that OpenAPPA reports no scored attack succeeded in the 1,320 guarded evaluations it describes.
What’s actually slowing this PC down?
Pick the symptom - the matching free tool is one click away.
Quick Recap
Best Value
- Used Book in Good Condition
Product prices and availability are accurate as of the date/time indicated and are subject to change. Any price and availability information displayed on Amazon at the time of purchase will apply.
