Are we ready for the AI fraud crisis?
A conversation about Sam Altman’s comments to the Federal Reserve, why document fraud is so hard to detect, and how agentic systems promise a new era of fraud prevention.
In this episode of Good Question, host Brianna Valleskey sits down with Inscribe AI co-founders:
- Ronan Burke, CEO, who’s spent nearly a decade working with financial institutions on the front lines of fraud prevention.
- Conor Burke, CTO, the architect behind Inscribe’s AI systems and a leading thinker on making machines actually helpful, not just powerful.
Together, they unpack OpenAI CEO Sam Altman’s warning of an “impending significant fraud crisis” and what it means for financial services, risk teams, and the future of trust.
Fraud in the age of AI: Volume, velocity, and variety
Ronan noted that Altman’s statement echoed what risk teams have already been seeing: AI has increased the volume, speed, and sophistication of fraud.
It’s not just Photoshop edits anymore. Fraudsters are now generating entirely fabricated documents and transaction data.
That means more losses, heavier workloads for operations teams, and poorer customer experiences.
“It’s the beginning of a new era,” Ronan explained. “Whether we call it a crisis is up for debate — but more leaders need to start ringing the alarm bells.”
AI-generated documents on the rise
Conor revealed a striking stat: Inscribe has seen a 200% increase in AI-generated fraud detections in recent months.
These fakes often slip past even manual review teams before being caught — leaving behind subtle clues such as:
- Uniform image sizes from common AI tools
- Dummy placeholder data
- Metadata artifacts and hidden watermarks
These markers provide an initial line of defense, but as fraudsters improve, detectors will need to evolve just as quickly.
From supervised models to fraud reasoning
The episode also traced Inscribe’s evolution in fraud detection:
- 2018–2020: Rules-based detectors and supervised ML models trained on specific fraud types.
- 2021–2023: Large language models for parsing documents, summarizing findings, and cross-document checks.
- Now: Moving toward agentic “fraud reasoning” systems that plan, use tools, and adapt like human investigators.
Instead of a one-to-one approach (one detector per fraud type), fraud reasoning enables a one-to-many model capable of catching both known and unseen fraud patterns.
“Fraud reasoning means we can finally tackle the long tail of fraud — the kinds we’ve never seen before,” Conor explained.
Why documents are the ultimate challenge
Documents remain among the hardest artifacts of trust to defend. They vary widely, mix visual and textual signals, and often require contextual reasoning.
An LLM-powered agent can ask:
- Does this pay stub align with the applicant’s history?
- Do the statement’s transaction dates make sense within the listed period?
- Is the metadata consistent with the claimed source?
By embedding reasoning in document authentication, fraud teams can protect one of the most important (and most abused) trust signals in financial services.
Inspiration from other hard sciences
Agentic systems are already making breakthroughs outside of fraud. Physicists recently used AI not to label data, but to discover new physical laws with messy, real-world inputs.
Bri drew a parallel: fraud is messy, nonlinear, and constantly evolving. Like in science, reasoning systems can uncover patterns no one has explicitly trained them to find, a powerful shift for fraud detection.
Document Fraudster of the Month: John Dixon
The episode closed with a real-world case study: John Dixon, former head of tax at EY, forged six declarations of trust and a loan agreement during bankruptcy proceedings in 2004. He simply downloaded templates online and added a law firm’s branding — a scheme exposed by unusual language and suspicious timing.
Imagine, the hosts pointed out, how much more convincing such forgeries would be with today’s generative AI tools.
A new vision for fraud prevention
This episode made one thing clear: fraud prevention is entering a new era. It’s not just about faster detection. It’s about smarter reasoning.
The future belongs to teams that embrace AI not as a replacement, but as a partner. Agentic fraud reasoning systems can spot patterns, adapt to new threats, and protect trust in ways that were impossible before.
“We’re embarking on a new way of building fraud detection systems; ones that can catch emergent patterns we’ve never seen before.”
👀 Watch the full episode on YouTube or visit the Good Question playlist.
🎧 Listen on Spotify, Apple Podcasts, or right here on this page. .
📢 Know a fraud fighter whose story should be featured? Let us know at info@inscribe.ai.
Sources Cited
- AI reveals unexpected new physics in dusty plasma
- Ex-Head of Tax at EY caught fabricating trust documents and calls it ‘normal practice’
About the guests
Brianna Valleskey is the Head of Marketing at Inscribe AI. A former journalist and longtime B2B marketing leader, Brianna is the creator and host of Good Question, where she brings together experts at the intersection of fraud, fintech, and AI. She’s passionate about making technical topics accessible and inspiring the next generation of risk leaders, and was named 2022 Experimental Marketer of the Year and one of the 2023 Top 50 Woman in Content.
Ronan Burke is the co-founder and CEO of Inscribe. He founded Inscribe with his twin after they experienced the challenges of manual review operations and over-burdened risk teams at national banks and fast-growing fintechs. So they set out to alleviate those challenges by deploying safe, scalable, and reliable AI.
Conor Burke is the co-founder and CTO of Inscribe. He founded Inscribe with his twin after they experienced the challenges of manual review operations and over-burdened risk teams at national banks and fast-growing fintechs. So they set out to alleviate those challenges by deploying safe, scalable, and reliable AI.
