Sponsored by GuardSquare (guardsquare.com), the discussion of Episode 333 opens with an analysis of a 1Password academic paper evaluating how frontier LLMs perform at autonomous vulnerability patching. The research indicates that LLMs successfully generate functional, side-effect-free patches only 26% of the time, often introducing new security flaws, breaking application behavior, or hallucinating fixes due to a lack of environmental context and "correctness collapse". The hosts critique the industry push toward auto-remediation, arguing that automated patch generation fails to address root causes like noisy tooling or organizational culture issues, and they emphasize that human domain expertise remains necessary for reliable patching. Turning to real-world AI security risks, the episode examines a Snowflake vulnerability where an AI coding tool (GitHub Copilot Autofix) regressed a GitHub Actions workflow into an unauthenticated Remote Code Execution (RCE) flaw via command injection, which was subsequently discovered and validated within five days by Wiz's automated "Red Agent". Finally, the hosts cover Dark Reading reporting on how the AI-driven "vulnpocalypse" is repricing the bug bounty economy. As automated scanning harnesses double report volumes, companies face budget constraints that reduce payout amounts per finding, forcing organizations to narrow program scopes toward high-priority assets.

In episode 332, the discussion focuses on how artificial intelligence is reshaping the Software Development Lifecycle (SDLC). The episode analyzes Anthropic's blog post regarding an "AI-native SDLC," evaluating its vision of replacing traditional development bottlenecks with AI workflows. The commentary critiques Anthropic's reliance on simple Markdown files for tracking development decisions, noting that replacing deterministic tools with probabilistic LLMs in core SDLC processes introduces significant reliability risks, context drift, and excessive token costs. The conversation turns to OpenAI's "Collective Call for Cyber Defense" initiative, examining its push for frontier AI model regulation and critiques of open-weight models, which are viewed as an effort to establish vendor lock-in. Exploring the concept of "Rumor as the Exploit," the discussion highlights how public mentions or minor disclosures of vulnerabilities now allow AI-driven testing harnesses to rapidly discover and generate working exploits across unmaintained software ecosystems. To counter this accelerated threat landscape, the episode evaluates defensive strategies, including runtime verification, reachability analysis, and cooling-off periods for new package releases, emphasizing that security defenders must move beyond thin wrapper solutions and build robust systems combining deterministic controls with model capabilities. Episode sponsored by Guardsquare (guardsquare.com).

Sponsored by Guardsquare (guardsquare.com), Episode 331 focuses heavily on the growing role of AI agents in application security and how organizations should build and defend against agentic systems. Ken and Seth argue that effective AI security systems should combine deterministic tooling with the probabilistic reasoning of LLMs rather than handing an entire security workflow to a model. Deterministic steps can map repositories, identify dependencies, reconstruct code relationships, and narrow the areas requiring investigation, while LLMs provide reasoning and creativity where those capabilities add value. Preparing for AI-assisted attackers, emphasizing secure development practices, guardrails, sandboxing, pre-production testing, and faster detection and response. The episode also examines HTTP request smuggling and CRLF-based attacks, including how differences in request parsing between proxies and backend services can create authorization bypasses and other exploit chains. Seth and Ken emphasize identifying the critical vulnerability within an exploit chain and discuss how service-oriented architectures can increase risk when components interpret the same request differently. Finally, they question whether bug bounty programs adequately reward researchers for discovering complex, high-impact vulnerabilities, especially as AI agents increasingly automate vulnerability discovery.

In this special episode of Absolute AppSec, we cover a topic which started as a solution proposed by Rippling Security's Jeevan Singh: Vulnerability Jail. As Jeevan describes it: "In this new AI world, we have seen many more vulnerabilities, and we struggled to get Engineering to fix them all in a timely fashion. This changed when we created Vulnerability Jail. If any vulnerability goes over SLA, your team is placed in Jail, preventing them from merging PRs into the main/default branches. We implemented Vulnerability Jail, updated our SLAs and got buy-in from Eng Leadership for our new Vulnerability Management program. As a result, we have now fixed the same number of vulnerabilities in one month as the team did in the previous year. The speed is still accelerating."" What do we think of the Stick approach to Vuln Management? What are the solutions to the rapid production by AI as well as the vulnpocalypse, Alex Gaynor's term to describe the way that "new technological innovation enables (or indirectly results in) finding a very large number of vulnerabilities in pre-existing software, which renders all previous assumptions and beliefs about the volume of extant vulnerabilities incorrect." Will "Vulnerability Jail" save us from the vulnpocalypse? Episode sponsored by GuardSquare (guardsquare.com)

In this episode, sponsored by GuardSquare (guardsquare.com), Ken Johnson and Seth Law discuss OpenAI's reported Hugging Face security incident, questioning whether the model demonstrated genuinely novel offensive capability or mostly chained known vulnerability patterns at high inference cost, while also considering the defense-contract and marketing angles around "dangerous" frontier models. The main technical discussion returns to AppSec fundamentals through an article on preventing IDOR, emphasizing authorization as a core control, the difficulty of role and tenant isolation in complex systems, and the need for framework-level patterns, typed IDs, tenant checks, and thorough authorization testing. They also cover Krebs' reporting on LG banning residential proxy SDKs from smart TV apps, explaining how free TV apps can turn consumer devices into proxy infrastructure and why IoT app ecosystems need stronger review. The episode closes with DEF CON logistics, Hacker Tracker updates, and upcoming guest plans.

In episode 328 of Absolute AppSec, sponsored by GuardSquare (guardsquare.com), Seth and Ken start by highlighting a newly disclosed, pre-authentication WordPress core Remote Code Execution (RCE) vulnerability ("WP2Shell"). The core discussion centers on Alex Gaynor's article regarding the influx of AI-assisted vulnerability disclosures. Gaynor and the hosts argue that attempting to fix bugs case-by-case is a "fool's errand"; instead, engineering teams must eradicate entire vulnerability classes through systemic, framework-level safe functions (such as parameterized queries) and automated CI/CD guardrails. They dive into the complexities of bug prioritization—debating reachability analysis, runtime verification, and business asset criticality—while noting that metrics and measurement remain among the lowest-scoring activities in OWASP SAMM assessments. Later, Ken and Seth examine a "Memory Heist" attack on Claude AI where indirect prompt injection tricked the assistant into exfiltrating user memory and corporate details letter-by-letter through web navigation. They conclude that because transformer models were originally designed for next-token prediction rather than secure system boundaries, defending LLM architectures behaves more like stopping social engineering than traditional software fuzzing.

In episode 327 of Absolute AppSec, co-hosts Ken Johnson and Seth Law present a highly anticipated quarterly crossover episode with Cameron and Kurt from the Coffee, Chaos, and ProdSec podcast. Sponsored by GuardSquare, the group begins with lighthearted banter about their personal footwear choices before tackling heavy architectural debates. The primary focus shifts to Application Security Posture Management (ASPM) consolidation. Cameron strongly advocates for utilizing ASPM as a distinct, single pane of glass dashboard to deduplicate vulnerabilities and streamline executive reporting by product suite. However, the hosts contrast this ideal against the messy reality of organizations dealing with a "Frankenstein" mix of loosely bootstrapped open-source scanning tools and competing vendor plugins.

The discussion deepens into prioritization strategies amid a massive, AI-driven surge in vulnerability research that threatens to double annual CVE counts. Cameron and Kurt stress the necessity of shifting away from abstract CVSS scores toward custom, runtime-informed risk appetites and impact analysis—prioritizing the hardening of high-risk corporate assets over low-reachability internal flaws. They also examine the critical line separating standard software bugs from intentionally malicious open-source packages that target developer endpoint systems. Ultimately, the panel laments that AppSec teams are effectively functioning as corporate incident responders because Security Operations Center (SOC) analysts lack product-level insight. The episode concludes with a review of automated agent statistics and a fun look ahead to the future emergence of meta OWASP top-ten risk lists.

In episode 326 of Absolute AppSec, sponsored by mobile application security provider GuardSquare (guardsquare.com), the hosts start with a deep-dive into pre-show discussions about the shifting macroeconomic landscape of AppSec jobs. They analyze an industry-wide trend where corporate hiring is pivoting away from external third-party consultancies and contractors. Instead, maturing organizations are forming internal product security "tiger teams" and hiring dedicated security software engineers across general development lifecycles to handle the exponential volume of code generated by artificial intelligence. Turning to AI-driven engineering, they dissect a research paper tracking security vulnerability mitigations through large language model (LLM) feedback. The paper reveals a distinct degradation in code quality and an explosion of "false positives" or unreachable flaws after the fourth or fifth iteration due to compressed context windows and "context drift." Ken highlights his own grueling experience benchmarking AINative software. He heavily cautions that letting models self-score or automatically review code introduces dangerous biases, reinforcing the absolute baseline requirement for humans to critically audit all LLM outputs. Finally, they examine Open Web Docs' new web security guidelines community group, comparing its browser-centric standard party focus to OWASP's broader, audit-driven charter. They close by promoting an upcoming July podcast collaboration with Coffee, Chaos, and ProdSec.

In episode 325 of Absolute AppSec, co-hosts Ken Johnson and Seth Law first break down an informal guide to threat modeling, arguing that overly prescriptive frameworks like STRIDE induce a heavy cognitive load on developers. Instead, they advocate for simplified, creative questions to expose architectural gaps, citing a historical GitHub planning flaw where private repository images were left exposed on S3 by relying solely on URL obfuscation. They warn that while rapid development in 2026 pushes toward automated lifecycles, human oversight, critical logging, and constructive friction remain essential. Next, they dissect a research paper exploring the philosophical definition of a vulnerability, framing it as a system disposition arising from a fault that manifests as a failure only when environmental and attacker conditions are jointly met. This definition sparks a debate on whether a flaw must carry immediate risk to qualify as a vulnerability, particularly when evaluating modern AI challenges like system prompt disclosures or exposed deprecated API paths.

In episode 324 of Absolute AppSec, co-hosts Ken Johnson and Seth Law share a mix of security model critiques. Starting with industry dynamics, Ken recaps his recent presentation at OWASP Nova regarding the limits of human-scale AppSec, recounting a dramatic storm during the talk where patio chairs pelted the high-rise glass. The conversation pivots sharply to Anthropic being forced to pull its "Fable" and "Mythos" cybersecurity models offline due to government sanctions and fears surrounding unpreventable universal jailbreaks. Ken and Seth criticize the company's disingenuous "FUD-based" marketing, which falsely suggested that AI could entirely replace security practitioners. Seth reviews his own blog post regarding the "three-week demo trap", detailing critical, ignored requirements for AI products—such as evaluation, statistical reproducibility, and token cost economics—noting that executing enterprise testing via frontier models can easily exceed $5,000 a day. Transitioning back to fundamental baseline defense, the hosts dissect an article on bypassing Visual Studio Code extension blocks. They emphasize that since modern CDNs pull zipped extensions from distinct domains, blocking the main marketplace URL is completely ineffective. Consequently, they advocate for rigorous data classification, layered on-premise model hosting, and stricter boundary controls on developer endpoints to combat fast-evolving supply chain threats.