Cybercrime Magazine Podcast · Cybercrime Magazine

Life Of A Cybersecurity Czar. Your AI Vendor: Your New Attack Surface. Dr Eric Cole, Secure Anchor.

March 24, 2026·21 min·2 clips

Dr. Eric Cole reveals how AI vendors use your company's private data to train their experimental models.

1. Cybercrime Radio's Scott Schober interviews Dr. Eric Cole on why AI vendors have become a new attack surface for enterprise security. 2. Dr. Cole has over 30 years of network security experience and worked as a professional hacker for the CIA; he now runs Secure Anchor, a cybersecurity consulting firm. 3. The episode's core thesis is that AI is no longer just a productivity tool — it is a potential backdoor into organizations, driven by the 'invisible vendor problem' of AI embedded across enterprise software. 4. Cole explains that established software vendors are embedding experimental AI into stable products and using customer data as training input, effectively making clients 'the training model' for evolving AI systems. 5. He raises the question of whether a company's private business data could be used to improve a competitor's AI system via shared vendor training data, framing it as an underappreciated data-management risk. 6. Cole invokes SolarWinds as an architectural warning: products that require 'access to everything' on a network create catastrophic risk if compromised, and AI systems are trending in the same direction. 7. On shadow AI, Cole states that boards are unlikely to treat it as a board-level risk while they are seeing $30 million revenue increases from unrestricted AI use — a direct tension with CISO recommendations to restrict access. 8. He reframes the CISO role as 'COIS' (Chief Officer of Information Security), arguing that practitioners must present revenue-positive solutions rather than restrictions, or they will be ignored. 9. His analogy: instead of 'jumping in front of a moving train,' CISOs should 'get on the train and into the engine room' — work within the AI adoption and then steer it toward security. 10. Cole explains that traditional third-party risk assessments evaluate vendor code pre-deployment but miss post-deployment plugin integrations that give a single AI system access to 20 previously siloed databases. 11. He argues that once private enterprise data is fed into public AI models, the distinction between public and private data becomes permanently blurred — a line he says 'nobody is adequately addressing.' 12. Cole's most alarming claim: privacy is 'a train leaving the station and never coming back,' and the next generation will have no functional concept of data privacy, relying instead on physical biometrics or chip-based identifiers. 13. He notes that social security numbers were originally unique identifiers, not authentication factors, and were misused as passwords — a design failure that required adding authentication layers, not replacing the identifier. 14. On AI data integrity, Cole argues that nation-states such as Iran are not attacking American infrastructure directly but are instead polluting the AI data sets that American organizations use for decision-making. 15. He distinguishes AI data poisoning from social media disinformation: social media targeted information, whereas AI poisoning corrupts the decision-making layer, which he calls a more dangerous form of influence. 16. Cole traces this risk to a failure in AI training methodology: when he built government AI systems in the 1990s to track terrorist activity, most of his time was spent acquiring accurate training data — a discipline now largely abandoned. 17. On AI risk governance, Cole quips that nobody clearly owns AI risk — 'the dolphins in the ocean' — because vendors don't communicate dynamic risk changes to enterprises and enterprises assume purchased products are secure by default. 18. He argues risk profiles for AI products are inherently dynamic, not static, because risk changes with usage volume and data ingested, requiring monthly risk analysis rather than annual assessments. 19. The interview is a conversational expert Q&A, with Schober providing questions and Cole delivering extended, opinion-driven answers drawing on his government and enterprise consulting background. 20. Security professionals and CISOs seeking a practitioner-level argument for reframing AI governance as a revenue opportunity rather than a compliance burden would find this most useful; listeners wanting technical implementation detail would not.

As heard by us

A sober warning that AI vendors can turn convenience into a larger attack surface.

The episode keeps coming back to one plain concern: AI turns trusted software into a bigger place to get burned. Dr. Eric Cole and Scott Schober treat the vendor stack as part of the attack surface, and Cole makes the point in practical terms.

Read the full review in PlayNext →

Why you'd press play

When AI tools become a backdoor into your company, you need this reality check.

Read the full recommendation in PlayNext →

Listen to the show on

Apple Podcasts Spotify

Play clips from this episode More from Cybercrime Magazine Podcast