The Pulse
Socure Reports 99.99% Lab Accuracy Identifying Muse and Instinct
Socure says its Agent Detection system identified the consumer AI agents Muse and Instinct with 99.99% accuracy across 700 behavioral variables in controlled testing. The company has made detection generally available, opened Agent TrustSco
AI.info Team · Written with AI · fact-checked ·
Socure says its new agent-detection system identified the consumer AI agents Muse and Instinct with 99.99% accuracy in controlled tests across 700 behavioral variables. The company announced Agent Intelligence on October 27, 2026, a set of tools designed to identify AI agents, assess whether their actions should be trusted, and connect them to the person or business that authorized them.
Three checks, with different release dates
Agent Intelligence combines Agent Detection, Agent TrustScore and Agent Credential. Detection is generally available through Socure’s Digital Intelligence product, and existing customers can turn it on immediately. TrustScore is open to beta customers, while Agent Credential is scheduled for general availability in the first quarter of 2027.
The tools address separate parts of an agent interaction. Detection identifies and classifies AI agents in browser sessions, including agents that do not identify themselves, and returns confidence scores and reason codes rather than a simple yes-or-no verdict. Socure says the system accounts for sessions in which control shifts between a person and an agent.
Scoring the agent and its principal
TrustScore assesses identity, authorization and risk as separate categories. Identity addresses who the agent is and who stands behind it; authorization concerns what it is permitted to do; risk draws on the histories of the principal and operator, including indications of replay or impersonation. Socure says customers receive the categories separately and can use them to set their own responses.
The credential component is designed to bind an agent to a verified person or business and document delegated authority. Socure says operators will issue it through SocurePass as a W3C Verifiable Credential. The format supports verification by a relying website without contacting Socure, as well as cryptographic chaining and proof of possession when one agent delegates work to another.
Socure’s case against treating agents as bots
Socure argues that existing fraud controls were built to distinguish people from bots, not to decide whether an automated actor is authorized to act for a person or business. Blocking all automated traffic can also block legitimate agents; accepting it without checks can leave businesses exposed to misuse. Johnny Ayers, Socure’s founder and CEO, described the trade-off in the company’s announcement:
“Block them all and you block your next customer. Let them all in and you hand fraudsters the keys, with little to no controls. The way out isn’t better bot detection. It’s clearly determining if it is an agent, which agent it is, and which human is on the other end of the agent, an evolution of the same problems we’ve been solving at Socure for more than a decade.”
Johnny Ayers, Founder and CEO of Socure
The system runs within Socure’s RiskOS platform and links agents to the company’s identity graph. Socure says that graph supports more than 11 billion identity decisions annually. The company also says TrustScore can assess agents without credentials and can evaluate credentials from issuers other than Socure.
The first test is outside the lab
The 99.99% figure is a company-reported result from controlled testing of named agents, not evidence that the system will achieve the same result across different agents or real-world traffic. The announcement does not give a test sample size or report results from customer deployments. Those details matter as organizations decide whether to rely on an agent’s score, its credential, or both.
Detection is available now, TrustScore is in beta, and credentials are not scheduled for general availability until Q1 2027. Evidence from live use and the acceptance of portable credentials by websites will provide further indications of how the system performs beyond controlled testing.