This week on Episode 336 of Absolute AppSec, Seth and Ken (@sethlaw and @cktricky, respectively) dig into Apple’s hardware-level photo signing, drone strikes on a Russian data center, and the decades-old vulnerabilities that AI coding tools are bringing back. To find all our episodes, head over to https://www.youtube.com/@AbsoluteAppSec/streams, or find us wherever you get your podcasts. Please be sure to leave us 5 stars on Apple Podcasts or Spotify. 

Before we go deeper (and to address the elephant in the room), we hope everyone made it through their php2shell response. This will be discussed soon, so watch out for some post mortem and thoughts on the current state of vulnerability identification and response. Keep your head up, stay sane, come share thoughts and experiences in Slack for the hard times when trying to deal with the current vulnpocalypse.

Psst! There’s still time to sign up for our course at Black Hat Europe, if you want to catch us outside of 9am Pacific/12pm Eastern— Come catch our AI-Enhanced AppSec training on December 7th and 8th. 

Ken Johnson and Seth Law opened with a story Seth had been waiting for: Apple now cryptographically signs photos at the hardware level, proving an image came from a real Apple camera instead of trying to detect whether it was AI-generated. Seth framed it in classic security terms, the allow-list beating the deny-list. Pattern-based AI detection, like the tools universities use to catch AI-written papers, is an arms race that new models keep winning, while a verifiable signature is far harder to dodge. Ken pointed out that Apple signs the metadata too, and said it’s no surprise Apple got there first, since it owns everything from silicon to software and every platform has a reason to stay on its good side rather than strip the signatures. Both see video calls and voice as the next places that need proof of a human. The thread led into the Ukrainian drone strike on a data center run by Yandex, Russia’s largest tech company, after which users noticed far less political bot activity in their feeds. Seth wants harder data, but Cloudflare’s figure of roughly 99.6 percent bot traffic from Yandex stunned them both.

Next was PortSwigger’s XSS cheat sheet, updated several times in 2026 with new bypasses. Seth’s point is that cross-site scripting never went away: Google still pays out heavily for it, and testers keep finding ways around deny-list filters. Ken thinks AI-assisted coding is bringing back vulnerabilities that should have died twenty years ago. Seth backed that up with a recent, possibly vibe-coded, client app where a basic ../../etc/passwd path traversal dumped the file, something he hadn’t seen outside a CTF in a decade. Both were candid about their own struggles. Ken has thrown out costly AI-drafted pull requests and says the code isn’t much better than a year ago, while Seth admits he accepts more Claude output than he reviews. Seth’s biggest complaint is with frontier labs whose guardrails reject legitimate security work even after he’s been verified. That pushes small firms toward open-weight models like Z.ai’s GLM, while attackers face no such limits. Both dismissed a widely shared “we have a year to secure everything” article as unrealistic, since most companies don’t even have an AppSec person on staff.

The episode closed on the state of the profession. Ken recalled interviewing candidates at GitHub about eight years ago and realizing a whole generation had learned AppSec as checking scanner output and chasing tickets, without reading code or breaking anything. He calls it being a “bug janitor” and says it wastes talented people, like treating a Ferrari as a bicycle. Seth agreed, noting that companies judge security by how many findings get closed and train employees mostly on phishing and passwords. Still, they wanted listeners to leave hopeful. AI isn’t the job-killer it’s been sold as, and Ken, who recently wrote a long essay in response to Anthropic’s AI development lifecycle playbook, believes software security will matter more as teams ship code they don’t fully understand. Curious people who see the world differently will catch what machines miss. The opportunity is to automate the triage and move into secure design, threat modeling, and advisory work. Seth added that good AI tooling surfaces fewer, more actionable findings, and clients need to stop expecting thousands of false positives.

Neiiiighhh. You can find it here:

If you’re looking for a community, come join us on our Slack!

Stay Secure,

Seth & Ken

Episode #281 — Signing Models, Vibe Coding, GitHub Action Abuse — The main topic of this episode is cryptographically signing machine-learning models with Sigstore, an open-source signing service. An interesting discussion, alongside Apple’s photo signing, on how it’s easier to prove where something came from than to detect AI mimics.

Episode #333 — LLM Patching Flaws, AI Code Regressions, Bug Bounty Economy — Just some more old bugs in new code. This episode covers a study finding that patches written by autonomous LLMs often fail, and a case where a Copilot Autofix change introduced a remote code execution bug. It also looks at how the flood of AI-generated reports is changing bug bounty payouts.

Episode #335 w/ Ron Perris — Secure Coding with AI — Listeners will remember Seth mentioning this conversation directly in 336. Rom talks about giving LLMs secure-coding guidance up front and about dashboards that judge security programs by how many vulnerabilities they count. 

Anthropic Reimagined the SDLC. AppSec Got Markdown Files — By way of DryRun Security, Ken (yes, that Ken) argues that Anthropic’s AI-native development playbook rethinks most of engineering but mostly just adds AI tools to existing AppSec work. He calls for security agents that handle threat modeling, code review and triage, with humans running those systems and stepping in on high-risk calls. He also wants teams to judge success by real risk rather than by counts of findings. 

Microsoft Breaks Patch Tuesday Record Again with 1400+ CVEs — Microsoft’s October 2026 Patch Tuesday fixed a record 1,411 vulnerabilities. Infosecurity Magazine credits the surge largely to Microsoft using AI to find and fix flaws. The release includes 70 critical bugs, and one actively exploited Windows Shell bypass. It also patches a no-click Outlook Reading Pane bug that a crafted email could use to run code, with no opening or clicking needed.

AI In Cybersecurity: Blizzard AppSec Chief Shares What Works And What Fails — Kurt Kaufmann, head of application security at Blizzard Entertainment, told CruiseCon attendees that AI helps most with mapping unfamiliar systems, confirming real vulnerabilities, and tracing many findings back to one shared cause. It fails when it completes tasks too literally, lacks company context or brings hidden costs. He urged teams to enforce limits outside the model, such as least privilege, approval gates, and kill switches, and to judge AI by verified findings rather than raw counts. 

Where in the world are Seth and Ken?

SaintCon - October 27-30, 2026 - AppSec Community, Speaking - https://saintcon.org/

Black Hat Europe 2026 - December 7-8, 2026 - AI-Enhanced AppSec: Black Hat Edition - https://blackhat.com/