The News
Arkose Labs, which positions itself as the leader in agent trust and orchestration, published its State of Agentic AI Trust Report, Q3 2026 on October 20, 2026. The report draws on production traffic data from July through September 2026 across banking, gaming, technology, and travel customers. Its central finding: AI agents are now actively transacting across the full customer journey, not merely browsing, and the controls businesses have in place are failing to keep pace. More than 65% of Arkose Labs’ customer sites observed agents logging in, signing up, and completing checkouts on behalf of end users, while fewer than 3% of those agents proactively disclosed their identity.
Analyst Take
The disclosure gap is the real story
The headline numbers in this report are striking, but the one that should concern security and product leaders most is the simplest: 97.7% of legitimate AI agent sessions carried no verified disclosure. These aren’t malicious actors hiding their tracks. They’re consumer agents, including Meta’s Muse, which appears on networks as a standard Chrome browser, operating transparently by their own design while remaining completely opaque to the systems they’re interacting with. This isn’t a fraud problem in the traditional sense. It’s an identity problem at a structural level, and existing bot-detection frameworks weren’t built for it.
The Arkose data makes a subtle but important architectural point. Treating agent traffic the same as human traffic creates revenue risk; blocking it wholesale does the same. The report notes that 70.5% of registration sessions were fake-account attempts, meaning the registration flow is already a high-stakes battleground. Layering legitimate AI agent traffic on top of that, without reliable disclosure signals, makes accurate classification exponentially harder. Businesses that apply legacy “allow all” or “block all” logic to this new traffic class will mis-route significant volumes of legitimate, revenue-generating sessions.
Intent over appearance: a new detection paradigm
The Muse growth curve, more than 30x in a single month, is a data point vendors in the fraud and authentication space should be studying carefully. It signals that personal agent adoption isn’t a gradual ramp; it can spike. The report’s finding that 96.8% of sessions from the Instinct agent looked medium or high risk, despite zero sessions being linked to an actual attack, exposes the core weakness of appearance-based detection. Risk scoring built on behavioral signals associated with bad actors will systematically misclassify well-designed consumer agents that happen to use similar techniques.
The practical implication for developers building on authentication and fraud-detection stacks is that intent inference needs to become a first-class capability. Behavior detection, biometrics, and network-level intelligence are the tools Arkose cites, but the broader market will need to converge on something closer to a declared-agent protocol or a widely adopted disclosure standard before intent can be reliably inferred at scale. Right now, that standard doesn’t exist, and the industry is operating on heuristics.
What this means for enterprise security buyers
For ITDMs, the business case for investing in agent-aware security infrastructure is becoming concrete. The 77.4% legitimate classification rate for AI agent sessions versus 7.9% for known bots tells you that the agent population isn’t a fringe phenomenon to be filtered out. It’s a growing segment of real customer interactions that carries revenue value. Getting the classification wrong in either direction has a cost.
This connects directly to a broader challenge in the public and regulated-sector context. According to ECI Research’s Google GovTech Survey, 31.8% of respondents identified FedRAMP/compliance approval friction for AI vendors as the single largest blocker preventing widespread AI adoption in developer workflows. That friction compounds when the agents themselves don’t self-identify, because compliance frameworks assume human actors or explicitly declared automated systems. An agent that presents as a Chrome browser is neither, and current Authorization to Operate models have no clean category for it. ECI Research’s same survey found that 46.2% of respondents reported their cATO initiatives are in progress, building automated compliance guardrails into pipelines. Those guardrails will need to account for non-disclosing agents before they can be considered complete.
The competitive dynamics in the agent trust space are also worth watching. Arkose Labs has built its position on a decade of bot-traffic intelligence, and that dataset is a genuine moat. Newer entrants will struggle to replicate the behavioral baselines needed to distinguish a well-designed consumer agent from a sophisticated bot without comparable production traffic to train against.
Looking Ahead
The Muse growth trajectory is a preview of what happens when a major consumer platform ships a capable personal agent to a large user base. Meta is not the only company with that kind of distribution. As Apple, Google, and others deepen agentic capabilities in their own ecosystems, the volume and variety of agent traffic hitting enterprise and consumer applications will compound quickly. The disclosure problem won’t self-resolve; if anything, competitive pressure among agent developers to minimize friction for end users creates an incentive to avoid disclosure flows that interrupt the user experience.
Over the next 12–18 months, expect regulatory attention to catch up with this gap. The EU’s AI Act and U.S. executive-level AI governance frameworks are already in motion, and agent disclosure requirements are a natural extension of existing transparency mandates. Organizations that build agent-aware detection and policy infrastructure now will be better positioned to demonstrate compliance when those requirements land, rather than scrambling to retrofit controls after the fact. For security vendors, the window to establish category leadership in agent trust is open, but it won’t stay open long once hyperscalers decide this capability belongs in their own platforms.
Stay Ahead of Application Development Trends
Get weekly analyst insights, research notes, event coverage, and AppDevANGLE updates delivered directly to your inbox.
Subscribe for Weekly Insights
Join technology leaders, practitioners, and GTM teams following the trends shaping modern software delivery.
Looking for deeper research access?
Explore ECI Research reports, survey insights, and market analysis through the ECI Research Portal.
