Pitched as the future of work, agentic AI is not selling well – but fraudsters love it

Everyone wants to talk about agentic AI. Throughout 2025, AI agents have been hailed as both the future of work and the new face of fraud. Predictions for 2026 continue to highlight the technology’s potential. According to the IBM Institute for Business Value, “agentic AI is emerging as one of the main tools leaders expect to use in the coming year, and most execs say AI agents are already helping them.” A new study from Juniper Research finds that the number of customer interactions automated by AI agents will grow from 3.3 billion interactions in 2025 to more than 34 billion by 2027.

In its list of the “8 biggest AI trends” to be ready for in 2026, Forbes predicts that agents will come to infiltrate everyday life (“it can take on day-to-day tasks like ordering groceries”) and says AI and automation represent “the future of work.”

And yet.

Doubts are beginning to creep in, as promises accumulate and ROI remains elusive. A piece published by Futurum Group asks, “Was 2025 Really the Year of Agentic AI, or Just More Agentic Hype?” And the latest report from identity verification and fraud prevention firm Au10tix, “Signals for 2026: The Au10tix Global Identity Fraud Report Special Edition,” goes so far as to suggest that 2025 was not the year of agentic AI – but rather, the “Year of Machine Deception, a turning point where synthetic identities and automated fraud engines learn, adjust, and adapt in real time.”

Time for agentic AI to deliver results: Futurum Group

Analysis from Futurum Group gives a quarter-by-quarter breakdown of the year, and comes to this conclusion:

“As the year draws to a close, enterprise software vendors continue to converge on a common strategic theme: agent platforms, supported by unified data and governed deployment, will define competitive advantage across the near- and medium-term. The winners will be those that pair robust multi-agent architectures with clear economic models and demonstrable customer outcomes and bottom-line value. While there has been significant progress around agentic AI, vendors are still tasked with making the case in 2026 that the technology can deliver true business value.”

In other words, agentic AI has to be worth it. Infinite agents that work for a business aren’t much good if they don’t improve efficiency or move the bottom line. The piece’s author, Keith Kirkpatrick, calls out the “claims wars” in which vendors compete to label themselves as the first or only tool to do a certain thing – rhetoric that can confuse the end user, but also bend expectations toward productivity that may or may not materialize.

‘The future that’s being sold to these customers simply hasn’t materialized’

Futurism has a damning report on AI agents, which suggests that the problem with all the proselytizing about how great agents are is that, at the moment, they’re not. “Researchers at Carnegie Mellon University found earlier this year that even the best-performing AI agent, which was Google’s Gemini 2.5 Pro at the time, failed to complete real-world office tasks 70 percent of the time,” the piece says.

The bluster is starting to cost companies money. Microsoft is reportedly struggling to sell its enterprise clients on its agentic AI scheme, and the firm’s stock dropped more than 2.5 percent last week, “showing that investors are unimpressed by the tepid sales results.”

“It’s yet another indication that companies are struggling to convert the enormous hype surrounding generative AI into actual revenue, a concerning trend considering the billions of dollars AI companies are burning through right now with no end – or return on investment – in sight.”

In short, says the piece, “the future that’s being sold to these customers simply hasn’t materialized. And that could hamper AI companies’ sky-high expectations when it comes to monetizing the tech.”

Defraud, rinse, repeat: fraud comes alive through agentic networks

Big firms are having a hard time finding things to do with their AI agents, but fraudsters are not. Au10tix’s report notes how agentic systems have turned fraud into something like a living entity:  “Fraud has entered a biological rhythm,” it says. “Attempt, analyze, adjust, repeat – where deception behaves less like code and more like an adaptive organism.”

The report continues Au10tix’s ongoing analysis of how fraud has “shifted from isolated attempts to adaptive, self-optimizing systems.”

“What began as ‘repeaters,’ recycled fraud behaviors that appeared across platforms, over time revealed distinct patterns that, once connected across industries, showed that seemingly isolated incidents were actually rehearsals for more adaptive attacks. These repeated signals have evolved into coordinated ecosystems powered by Fraud-as-a-Service and automated feedback loops.”

The firm says the pattern is made possible by three factors: open-source AI models and APIs that are simple to access and use; automation frameworks which allowed those models to chain actions across systems without human oversight; and feedback data from years of failed verification attempts that have become “massive training sets for optimizing synthetic behavior.”

Au10tix’s predictive assurance tool is “designed to detect convergence rather than consequence” by continuously analyzing behavioral, biometric and metadata signals across billions of identity events. “Each data point is scored, cross-referenced, and contextualized within a shared intelligence layer that connects what no single transaction could reveal.” Per the report, its early detection method “demonstrated a 97.5 correlation between early anomalies and confirmed fraud attempts.”

Regardless, there is an emerging case to be made that AI agents are best used as a tool for fraud, and that in pushing them as the future of work, businesses have bitten off a rather chewy noodle.

Multifactor offers secure online account sharing for authentication

For now, the money keeps rolling in. “Post-quantum security platform” Multifactor recently raised 15 million dollars in seed funding, in a round led by Nexus Venture Partners, with participation from Y Combinator, Taurus Ventures, Honeystone Ventures, Flex Capital, Pioneer Fund, Ritual Capital, Liquid2 Ventures and more.

A release says Multifactor’s “unique approach replaces the need for traditional password managers, turning any online account into a shareable ‘read-only’ link in the same manner to how Google Docs can be shared for viewing or editing.”

Founded by former CIA officer Vivek Nair and mathematician and cryptographer Colin Roberts, Multifactor was built specifically in response to risks raised by agentic AI. “For AI agents,” it says, “Multifactor provides a provably safe execution environment that prevents prompt injection, password theft, confused-deputy attacks, and the growing category of cross-agent hijacking incidents. Access is entirely capability-based: agents can only see or modify what the user has granted in advance.”

Nair, the company’s CEO, throws down the gauntlet introducing his product. “People aren’t going to hand AI agents the keys to their calendars, finances, or business systems unless they have complete confidence nothing will go wrong. Multifactor is the easiest, safest, and most verifiable way to collaborate with both humans and AI with security guarantees that are mathematically unbreakable.”

Article Topics

AI agents  |  AU10TIX  |  digital identity  |  identity access management (IAM)  |  Multifactor