Expert Insights Podcast

Deepin Desai delivers several genuinely non-obvious points - agent-as-malicious-insider, shadow AI jumping from developer laptops into CI/CD, and indirect prompt injection targeting LLM agents - but the episode is diluted by a second guest (Martin Ditchburn) whose section is largely generic commentary and vendor positioning, pulling overall density down.

“Agents, when they fall for phishing, they will become malicious insiders”

“one of the developer installed OpenClaw and within a week they found OpenClaw instance in their CI CD environment and they had no clue how the agent jumped from developer machine to CICD”

The agent-phishing-agent prediction and the framing of compromised agents as 'malicious insiders who never report it' are fresh and counterintuitive; however, the bulk of the advice (zero trust, MFA, training, segmentation) is recycled standard cybersecurity doctrine, and Martin's section contributes almost no novel thinking.

“It's basically a type of attack where the threat actor is able to fish the agent into doing certain tasks that it was not originally programmed to do”

“the agent phishing agent. It sounds cool, but people are not understanding the gravity of it”

Deepin Desai is a legitimate senior practitioner (EVP/CSO) who clearly has hands-on threat intelligence and uses real CxO anecdotes, but the inherent vendor bias of a keynote speaker at his own company's conference limits credibility; Martin Ditchburn is a field/sales-adjacent CTO-in-Residence role whose contributions are largely vague market commentary.

“in one of the discussions I was having with the CxO, they said one of the developer installed OpenClaw and within a week they found OpenClaw instance in their CI CD environment”

“if I've done proper segmentation and I have these decoys planted in the environment, when the adversary uses that compromised identity or the asset to do more damage, they will get caught”

The Deepin segment is meaningfully specific - 440,000 AI-generated phishing pages, 95% over TLS, 121,000 throwaway cloud servers, named tools (Manus AI, Vercel, Lovable), named packages (LightLLM, Axios), and named campaigns (Payout Kings) - but Martin's segment is almost entirely free of data points or named examples, which drags the score down.

“we saw about 440,000 phishing pages that were created using AI, and they were all leveraging these SaaS, service providers”

“about 75% of the victim Personas were either in uh, sales, HR and operations”

The host structures questions logically and threads findings from the threat report into follow-ups, but she consistently validates rather than challenges ('Quality over quantity is a really good way to put it'), never presses on vendor self-interest or conflicting claims, and lets Martin's vague answers pass without probing for specifics.

“Quality over quantity is a really good way to put it. Just because there's less of them doesn't mean they're less of a threat.”

“I guess you can't explain expect the same loyalty from agents”