DiscoverAI Spy
AI Spy
Claim Ownership

AI Spy

Author: Anurag Mohapatra

Subscribed: 4Played: 17
Share

Description

Join us in the world of financial crime prevention, where we delve into the powerful synergy of Artificial Intelligence and cutting-edge technology. This podcast brings expert insights, no-nonsense discussions, and deep dives into topics like Anti-Money Laundering, Payments Fraud, and more. Hosted casually and informatively, we invite A-listers from the industry to share their experience and learnings.
37 Episodes
Reverse
Anurag talks with Dr. Rasha Kasim, a senior academic at Aston University, founder and lead of its fraud research group, and a certified fraud examiner. She received the 2025 ACFE Educator of the Year Award and sits on the UK government's Counterfraud Professional Cross-Sector Advisory Group and the ACFE Advisory Group. The episode covers how romance fraud works, why the language we use matters, who should share responsibility, and how AI is changing the crime.In this episodeHow the Enron collapse, which Rasha encountered while studying accounting, led her into fraud research, and how her work expanded from corporate fraud to fraud against individualsWhy romance fraud hits so hard: victims suffer financial and psychological harm, and Rasha's current research finds it among the most damaging fraud types for physical and mental healthWhy she argues for saying "fraud" instead of "scam": fraud has a legal definition, scam does not, some banks treat the two differently when reimbursing, and "scam" can imply victim blameWhy she disagrees with "pig butchering" as dehumanizing to victimsThe stages she describes: building a persona matched to the victim, love bombing, moving the victim off the platform, then manufactured crises that never endHarm beyond lost money: one victim was pushed into heavy debt through her credit cards, and another was used as a money mule without realizing itExtortion when victims push back, including sextortion, AI-altered images and threats of physical harmWhy anyone can be a victim regardless of age, gender, race or wealthShared responsibility across families, police, banks, dating and social platforms, higher education and governmentHow AI improves fake personas, polished messages, fake documents and cloned voices, and enables re-victimization, including recovery fraudResources: Dr Kassem's LinkedIn - https://www.linkedin.com/in/dr-rasha-kassem-phd-cfe-cpa-sfhea-cmbe-3ba82b40/https://www.aston.ac.uk/research/bss/research-centres/fraud-research-group/recent-publicationshttps://www.aston.ac.uk/research/bss/research-centres/fraud-research-group/fraud-guidance
This episode explores why traditional model risk governance does not fully fit agentic AI, especially when systems can choose their own path and act at machine speed. Anurag is joined by Vyasa Murthy, CEO and founder of Venture Vertex, who explains why banks and regulated organizations need real-time control, not just logs and after-the-fact review.We discuss how agent chains create new accountability gaps, why evidence must be captured at the moment of action, and how runtime governance can become a competitive advantage as AI moves deeper into critical workflows.Key topicsIn this episode, Anurag frames the shift from AI that recommends to AI that acts, and why that changes the governance problem in regulated environments.Vyasa Murthy explains his background in enterprise technology across banking, cards, wealth, and insurance, and how years of seeing controls quietly erode led him to build Venture Vertex and VARC.The conversation contrasts classic model risk management with agentic systems, where the approved behavior is no longer a fixed function but a path chosen at runtime.Vyasa argues that logging alone is not enough because by the time an analyst reviews the record, the damage may already be done.A key example is data loss prevention: a control may start with a legitimate exception, then slowly become effectively disabled for the sake of convenience.The episode shows how chained agents can each pass their own checks while the overall outcome, such as a payment, has no single accountable owner.Anurag and Vyasa connect this to segregation of duties and the human “uneasy feeling” that often acted as an informal guardrail in manual workflows.Vyasa describes VARC as verifiable AI runtime control, designed to intervene at the moment an agent tries to act, not after the fact.The discussion introduces graduated response orchestration, moving beyond simple allow or deny decisions to options like flag, escalate, freeze, or decommission a session.The episode closes on the idea that organizations can outsource components of the AI stack, but not accountability, evidence, or the ability to prove what their agents actually did.
In this episode, I sit down with Sune Selsbæk-Reitz, a historian whose work traces how confident, fluent language gets mistaken for reliable knowledge, a pattern he now studies in how we relate to AI. His book Promptism names and unpacks this behavior for a generative AI audience.We start with what Promptism actually means, and why a historian is the one writing about it right now. Sune walks through the Authority of Confidence, his idea that wrongness wrapped in certainty doesn't read as error, it reads as leadership. From there we get into his Delphi reframe: when the answer disappoints us, we've been trained to blame the prompt, never the tool. I bring in what this looks like on the ground in product management and marketing, from one-line-prompt PRDs to fact sheets written entirely by an LLM. We also talk about where AI genuinely earns its place, including Sune's own use of it as a search engine for his own thinking. We close on the argument from his RÆSON op-ed: that we're not just automating language, we're at risk of automating responsibility away.Links:Sune Selsbæk-Reitz LinkedIn — https://www.linkedin.com/in/suneselsbaekreitz/Sune's Substack - https://footnotesandfriction.substack.com/Promptism (book) — https://a.co/d/0eeLGN4w#Promptism #AI #Knowledge
The security community has been sounding alarms about AI infrastructure for two years. BadHost, a critical authentication bypass discovered in May 2026 inside one of the most widely used Python frameworks on the internet, is what that alarm sounds like when it goes off. Dan Fernandez returns to AI Spy to talk about what actually happened, why the official severity rating understates the real risk, and what the broader pattern of 40-plus CVEs against MCP implementations in 2026 alone tells us about how AI infrastructure is being built. This is not a technical deep-dive. It is a practitioner briefing on the vulnerability surface that every organization deploying agentic AI is now sitting on, whether they know it or not. Show NotesDan Fernandez, cybersecurity practitioner and returning AI Spy guest.Dan has spent his career at the intersection of threat intelligence and enterprise security infrastructure. He returns to AI Spy for his second appearance having previously covered AI as an attack tool. This time the conversation flips: AI infrastructure itself is the attack surface.In this episode:Vulnerability management 101 and what fundamentally changed once AI became load-bearing infrastructure inside organisations, not just a layer on top of themWhy BadHost (CVE-2026-48710), a flaw in the Starlette Python framework used by Netflix, Uber, Microsoft and hundreds of thousands of other projects, carries a severity rating that most practitioners consider a polite fictionHow MCP servers aggregate credentials across every service an AI agent is connected to, and why that makes a single compromised server a portfolio-level exposure eventClaude Mythos found over 10,000 vulnerabilities through Project Glasswing and missed this one. What that tells us about where AI-assisted security tooling is strong and where it has genuine blind spots40-plus CVEs against MCP implementations in 2026 alone. Whether BadHost is an outlier or a pattern in how the AI infrastructure layer is being builtWhat a defensible enterprise posture actually looks like for organisations deploying AI agents, AI-generated code, and MCP-connected systems right nowLinks and References: BadHost - CVE-2026-48710 Starlette Host-Header Auth BypassDisclosing the BADHOST Vulnerability in Starlette – OSTIF.orgDan Fernández | LinkedIn
Most of us in fraud prevention have never spoken to someone who has been inside a scam compound. Ling Li has. A Research Fellow at the University of Melbourne with a decade in law enforcement experience and field roles across the ICRC, UNHCR, and the University of Liverpool's Modern Slavery Research Centre, Ling co-authored SCAM: Inside Southeast Asia's Cybercrime Compounds — one of the most rigorous accounts of how these operations actually work. In this conversation, she traces the 30-year lineage of the industry from Taiwan to Fujian to militarized compounds across Cambodia and Myanmar, unpacks the victim-offender overlap that makes conventional law enforcement responses inadequate, and explains how AI has expanded the operational reach of these networks in three specific ways. She also raises an alarm that the current crackdown is generating a new crisis: thousands of survivors stranded without shelter or repatriation support, being recycled back into the same compounds. Fraud is a human problem. This episode is the evidence.In this episode:How Ling discovered the scam compound world in Cambodia in 2022 while researching bride trafficking — and why almost no international organisations were responding to Chinese victims at that timeThe 30-year arc from 1990s Taiwan phone fraud to Fujian scam towns to today's militarised, multi-continental operations — and why every crackdown relocates rather than eliminates the industryThe victim-offender overlap: four archetypes from pure criminal to fully coerced victim, why the majority fall in between, and what a proportionate law enforcement response looks likeHow AI has changed the operations: real-time multilingual translation, face-swapping that defeats video verification, and automated phishing site regeneration that outruns detectionThe post-crackdown crisis: thousands of survivors sleeping on streets across Southeast Asia, no food, no shelter, no repatriation — and being approached by traffickers offering the cycle againLinks:Ling Li on LinkedIn — LING LI | LinkedInSCAM: Inside Southeast Asia's Cybercrime Compounds, Verso Books — Amazon.com: Scam: Inside Southeast Asia's Cybercrime Compounds: 9781804296905: Franceschini, Ivan, Li, Ling, Bo, Mark: Books
loading
Comments