
Your AI provider can silently swap the model behind your request. Johann Rehberger: An attacker who forces a downgrade only has to beat the weaker model. Rerun your evals on every model change. #AppSec #AISecurity appsecpodcast.com/trust-no-a…

Christian • Husband • Dad • Granddad | CEO, Stealth • GP, @Kerr_Ventures | Host: @AppSecPodcast + @SecTablePodcast | AppSec • Threat Modeling • AI Security

Your AI provider can silently swap the model behind your request. Johann Rehberger: An attacker who forces a downgrade only has to beat the weaker model. Rerun your evals on every model change. #AppSec #AISecurity appsecpodcast.com/trust-no-a…

Johann Rehberger gave a Codex session one job: watch his GLM 5.3 agent's logs. It caught the agent trying to SSH to the gateway. Out of the blue. Why every agent needs a sentinel #AppSec #AIAgents youtube.com/watch?v=kzV_qgLa…

A human insider planting backdoors goes to jail. What happens to an AI agent that does it? #AISecurityTable #AIAgents

Johann Rehberger breaks AI agents for a living. He still won't run one autonomously on his own computer. His rule: trust no AI. #AppSec #AISecurity youtube.com/watch?v=o7v89dTO…

Let your agents work. Keep them in sight. Origin inventories, observes, and measures AI agent usage across endpoints so security teams can see what agents do. Thanks to Origin for sponsoring this week's episode with Rock Lambros. #AppSec #AIAgents appsecpodcast.com/origin.

Patch the code you understand, or regenerate the code you don't? Pick one. #AISecurityTable #AppSec securitytable.ai/stop-patchi…

"Make no mistakes" is not a pragma. LLMs are not compilers. Stop treating them like one. #AISecurityTable #AppSec #Shorts

Claude writes the code, the patch, and the proof. Does compounded mistrust ever equal trust? #AISecurityTable #AISecurity #Shorts

Teaching privacy works. People understand it. But as Avi Douglen puts it, "there's a world of difference between understanding it and getting it." That gap is why he and Kim Wuyts built Context & Cringe. #AppSec #Privacy appsecpodcast.com/the-cringe…

Regex was everything. Then ASTs. Now tokens. Are LLMs the real thing or the next oversold scanner? #AISecurityTable #AppSec #LLM securitytable.ai/you-don-t-t…

Altman: We can't just patch all the bugs. Izar: The team said "security debt" on top of that. We stood there anyway. #AISecurityTable #AppSec #Shorts

AI can draft your privacy threat model. It can't feel the cringe. Kim Wuyts and Avi Douglen on Context & Cringe, APVS, and why the human gut still matters in privacy. #AppSec #Privacy youtube.com/watch?v=ABKLYZhF…

Ever wanted to be a guest on The Application Security Podcast? Here's your chance. We're recording a live episode at Global AppSec in San Francisco, and this time the audience is part of the show. First, an Ask Me Anything. You bring the questions on AI and security, and we (plus a few special guests) answer them live. Then we flip it. Reverse AMA: we ask you the questions, and the audience becomes the guest. (Fair warning: we've been saving up some good ones.) Thursday, November 5 10:15 to 11:00 a.m. PT Regency Room Can't make it to SF? Drop your question in the comments. We might use it on the mic. #AppSec #OWASP #AISecurity #ApplicationSecurity

You don't trust AI. So why ship its code? Altman says we can't patch all the bugs. Claude is writing its own proofs. Who checks the checker? #AISecurityTable #AISecurity #AppSec youtube.com/watch?v=xs67gzLv…

Scion is infrastructure rebuilt for agents first, people second. When your tools are optimized for agents, who's still checking what the agents did? #AISecurityTable #AISecurity #AgenticAI securitytable.ai/anthropic-d…

Can you still trust the CVE process when the exploit shows up before the CVE does? Hannah Foxwell and Chris Romeo on why exploitability is the wrong thing to wait for. #AppSec #VulnerabilityManagement appsecpodcast.com/appsec-has…