Securing the Realm

The episode contains solid substantive content on agentic security, regulatory divergence, and commodification with specific examples (Singapore's January framework, UK AI Security Institute de-scoping, Comet browser exploit in 150 seconds, llama firewall), but meanders frequently and includes stretches of obvious or exploratory discussion that dilute the insights-per-minute ratio. The discussion about soft deletes and CRUD operations feels tangential.

“Singapore, um, I think one of the key things for them is around what does all this mean for their people... instead of seeing AI as LLMs, it does at with that agentic application lens.”

“Comet with this perplexity browser got Comet jacked in 150 seconds. So it's just over two minutes from a Reddit comment with an indirect prompt injection attack to Comet giving away credentials.”

The episode reframes AI security around agents rather than LLMs, discusses the OWASP agentic security lens, and explores less-traveled angles like ambient AI, observability-vs-visibility distinction, and the shift from models-as-building-blocks to architecture-as-competitive-advantage. However, the core frameworks (governance, regulation, commodification) are relatively well-trodden in 2026 discourse, and some takes (SaaS death, data layer primacy) are presented as novel but circulate widely.

“OWASP was the first agentic security piece that really focused on agents and not the large language models underneath.”

“There is no security through full visibility... It's too much data.”

Both speakers position themselves as MVPs in the Foundry space and ex-lawyers now working in agentic security, which suggests practitioner-level relevance. However, the transcript provides no concrete evidence of either speaker's specific scale, portfolio, or operational achievement. They discuss frameworks and trends rather than detailing their own implementations or lessons learned from large deployments, which would establish true operator caliber.

“We're both MVPs in the Foundry space.”

“We were both once lawyers.”

The episode includes several named examples (Singapore framework, Comet/Perplexity breach, llama firewall, Open Claw, MCP, A2A protocol, OpenAPI, LLM.txt, Claude Opus 1M token window) and one specific metric (llama firewall in the high nineties for injection detection, 150 seconds for Comet jacking), but lacks depth on implementation details, dollar figures, deployment scale, or concrete deployment metrics. Most claims about adoption, standards adoption, or security outcomes remain asserted rather than evidenced.

“Comet with this perplexity browser got Comet jacked in 150 seconds.”

“Meta has just got their llama firewall, which I think is in the high nineties for identifying prompt, prompt injection techniques”

The host-guest dynamic feels collegial and collaborative rather than adversarial or deeply probing. Questions are open-ended and exploratory (e.g., 'What do you think about that?'), and follow-ups exist but rarely challenge claims or dig into contradictions. When Josh introduces skepticism (e.g., 'I haven't tried agentic browsers'), Chris pivots rather than pressing. The conversation meanders thematically without tight argumentative pressure, which suits a roundtable but limits depth.

“What did think about that? Were you following that?”

“What you seeing in the rest of the world?”