Sign up to save your podcasts
Or
Share 96% Human Error: Why AI Security Starts with the Human, Not the Model
Share to email
Share to Facebook
Share to X
Share to email
Share to Facebook
Share to X
Or
96% Human Error: Why AI Security Starts with the Human, Not the Model
Some travel operators ask you to shout your passport number across a crowded desk and think nothing of it. While intentions are good (checking who you are), this episode is about why that is a serious security failure and what it would take to fix it.
Yagub Rahimov is the CEO and founder of Polygraf AI, a company building behavioral security and contextual privacy tools for enterprise environments. In this conversation, he and Alex work through the specific vulnerabilities created when AI agents gain user level access, why human behavior rather than model failure is responsible for the vast majority of data breaches, and what a genuinely privacy respecting travel product would actually look like.
What You'll Learn:
- Agent security: AI agents are a new category of user in the digital security pyramid, with the same system access as humans but no training in deception or social engineering.
- Deep fake risk: Voice cloning is already sophisticated enough to impersonate individuals convincingly to family members and colleagues, without any technical breach of the underlying systems.
- Mosaic intelligence: Even anonymized data fed repeatedly to an AI can be re-identified over time through behavioral pattern mapping, a concept Rahimov terms "mosaic intelligence."
- Behavioral control: Addressing human behavior in real time, before a violation occurs, is more effective than after-the-fact audit or punitive controls, demonstrated by a 72% drop in DLP violations for one enterprise client.
- Data in AI tools: Organizations that deploy internal LLMs without governing what employees input are creating serious exposure, as one $25M chatbot deployment illustrated on its first day.
- Travel industry failures: Asking passengers to recite passport numbers and dates of birth aloud in crowded gate areas, or type personal data into in-flight entertainment screens, represents a real and unaddressed privacy risk.
- Tokenization as a fix: Stripping personal data before it reaches an LLM and reuniting it with processed output via tokenization can deliver the same analytical value with substantially less exposure.
- QA at scale: AI makes universal quality assurance of customer interactions cheap enough that random sampling is no longer the only option, with one call center client processing 500,000 calls daily at 5 to 10 cents per call.
Time-Stamped Highlights:
- (00:00) Introduction: The airport data disclosure problem
- (00:00:42) What actually happened with Meta's Instagram AI chatbot
- (00:06:17) AI agents as a new user type: the security pyramid explained
- (00:08:57) Deep fakes in practice: voice cloning, elderly parents, and the CEO
- (00:14:36) North Korean infiltration via data science job interviews
- (00:20:54) How Polygraf detects synthetic speech in real-time video calls
- (00:28:42) The meeting note taker with 23 vulnerabilities
- (00:36:24) How Mr. Paranoid travels: loyalty status, one airline, mid-tier hotels
- (00:42:58) The oil and gas CEO kidnapping and the email summarizer attack vector
- (00:49:00) What travel companies get wrong about passenger data collection
- (00:30:10) Mosaic intelligence and why anonymizing data is not enough
- (01:07:07) The $25M HR chatbot and the 72% DLP violation reduction
- (01:14:12) Building the next OTA: tokenization, QA at scale, and simplicity
- (01:21:01) Red teaming, visibility, and why behavioral control is the next frontier
Guest bio: Yagub Rahimov is CEO and founder of Polygraf AI, a company specializing in behavioral security, contextual privacy, and AI risk management for enterprise clients. He works across defense, financial services, and enterprise technology sectors, and is an active contributor to conversations on AI behavioral control at venues including the Gartner Security Summit. LinkedIn: https://www.linkedin.com/in/yrahimov/ | Company: https://polygraf.ai/
About the Podcast: The Travel Tech Podcast features long-form conversations with leaders across travel and technology. The show explores how software, data, operations, and distribution come together in real businesses, with an emphasis on tradeoffs, incentives, and lessons that transfer beyond any single company or role.
Host bio: Alex Brooker is the founder of Airside Labs, an AI business that applies aviation-grade testing and compliance rigor to enterprise AI systems. Before founding Airside Labs, he built and scaled complex software in aviation and safety-critical domains, and he invests in early-stage technology ventures. LinkedIn: https://www.linkedin.com/in/alex-brooker-2280002/
View all episodes
By Alex BrookerSome travel operators ask you to shout your passport number across a crowded desk and think nothing of it. While intentions are good (checking who you are), this episode is about why that is a serious security failure and what it would take to fix it.
Yagub Rahimov is the CEO and founder of Polygraf AI, a company building behavioral security and contextual privacy tools for enterprise environments. In this conversation, he and Alex work through the specific vulnerabilities created when AI agents gain user level access, why human behavior rather than model failure is responsible for the vast majority of data breaches, and what a genuinely privacy respecting travel product would actually look like.
What You'll Learn:
- Agent security: AI agents are a new category of user in the digital security pyramid, with the same system access as humans but no training in deception or social engineering.
- Deep fake risk: Voice cloning is already sophisticated enough to impersonate individuals convincingly to family members and colleagues, without any technical breach of the underlying systems.
- Mosaic intelligence: Even anonymized data fed repeatedly to an AI can be re-identified over time through behavioral pattern mapping, a concept Rahimov terms "mosaic intelligence."
- Behavioral control: Addressing human behavior in real time, before a violation occurs, is more effective than after-the-fact audit or punitive controls, demonstrated by a 72% drop in DLP violations for one enterprise client.
- Data in AI tools: Organizations that deploy internal LLMs without governing what employees input are creating serious exposure, as one $25M chatbot deployment illustrated on its first day.
- Travel industry failures: Asking passengers to recite passport numbers and dates of birth aloud in crowded gate areas, or type personal data into in-flight entertainment screens, represents a real and unaddressed privacy risk.
- Tokenization as a fix: Stripping personal data before it reaches an LLM and reuniting it with processed output via tokenization can deliver the same analytical value with substantially less exposure.
- QA at scale: AI makes universal quality assurance of customer interactions cheap enough that random sampling is no longer the only option, with one call center client processing 500,000 calls daily at 5 to 10 cents per call.
Time-Stamped Highlights:
- (00:00) Introduction: The airport data disclosure problem
- (00:00:42) What actually happened with Meta's Instagram AI chatbot
- (00:06:17) AI agents as a new user type: the security pyramid explained
- (00:08:57) Deep fakes in practice: voice cloning, elderly parents, and the CEO
- (00:14:36) North Korean infiltration via data science job interviews
- (00:20:54) How Polygraf detects synthetic speech in real-time video calls
- (00:28:42) The meeting note taker with 23 vulnerabilities
- (00:36:24) How Mr. Paranoid travels: loyalty status, one airline, mid-tier hotels
- (00:42:58) The oil and gas CEO kidnapping and the email summarizer attack vector
- (00:49:00) What travel companies get wrong about passenger data collection
- (00:30:10) Mosaic intelligence and why anonymizing data is not enough
- (01:07:07) The $25M HR chatbot and the 72% DLP violation reduction
- (01:14:12) Building the next OTA: tokenization, QA at scale, and simplicity
- (01:21:01) Red teaming, visibility, and why behavioral control is the next frontier
Guest bio: Yagub Rahimov is CEO and founder of Polygraf AI, a company specializing in behavioral security, contextual privacy, and AI risk management for enterprise clients. He works across defense, financial services, and enterprise technology sectors, and is an active contributor to conversations on AI behavioral control at venues including the Gartner Security Summit. LinkedIn: https://www.linkedin.com/in/yrahimov/ | Company: https://polygraf.ai/
About the Podcast: The Travel Tech Podcast features long-form conversations with leaders across travel and technology. The show explores how software, data, operations, and distribution come together in real businesses, with an emphasis on tradeoffs, incentives, and lessons that transfer beyond any single company or role.
Host bio: Alex Brooker is the founder of Airside Labs, an AI business that applies aviation-grade testing and compliance rigor to enterprise AI systems. Before founding Airside Labs, he built and scaled complex software in aviation and safety-critical domains, and he invests in early-stage technology ventures. LinkedIn: https://www.linkedin.com/in/alex-brooker-2280002/