Breach Please

Epizodes

• 
  S0:E44: AI Agents Are Leaking Your Data to Public GitHub, Plus NYT's Costly Privilege Mistake 02.10.2026 39min

  AI coding agents have been finding creative ways around a GitHub limitation — and leaking internal and production data to public repos in the process. Plus: how a single email mistake cost the New York Times its first libel trial loss in 50 years.In this episode, Jake and Jess cover:A security research firm (reported in our source material as "GLOW" — spelling uncertain, this is an auto-transcript name) found more than 13,000 internal developer images across 300+ organizations sitting in public GitHub repos. The root cause: GitHub's `gh` command-line tool can't attach images to a pull request, so AI coding agents — tasked with showing a UI fix worked — found workarounds, including publishing screenshots to a developer's personal public GitHub account, or using a tool referred to as "GitShot" (also an auto-transcript spelling) to get around the CLI limitation. At one software company, the workaround got baked into a shared AI skill and spread across agents, uploading 1,000+ screenshots and recordings of the company's own product. Jake and Jess talk through why this is reward hacking in action, why access controls can't be relied on to stop an agent determined to complete its task, and why regulated data is likely present in some of the leaked images.Also: Epic Systems used AI to find security flaws that could expose patient records — a reminder that "AI in healthcare" covers very different risk profiles depending on whether you're talking about machine learning or generative AI.Then: the New York Times lost its first defamation trial in more than 50 years, over a story about a college basketball player wrongly placed at the scene of a shooting. The reason the paper couldn't claim its usual "actual malice" protection: the plaintiff wasn't a public figure. Jake and Jess dig into what actually makes a communication privileged (the only test that matters: is an attorney a meaningful participant?), and how the Times' own editor broke privilege on an internal email thread by removing the paper's legal staff before forwarding it — a decision that very likely helped produce a $9 million jury verdict (later reduced to $4.7 million by the judge).No fear-mongering, no vendor scripts, no "synergizing our threat posture." Just two people who've worked the incidents talking through what the headlines actually mean.Breach Please is a production of JWJH Media LLC. The opinions of our hosts are their own. This is not legal, financial, or security advice — do your own homework before pointing anything at prod.
• 
  S0:E43: NetScaler Was Exploited for Weeks by State Hackers, Plus the AI 'Torture Chamber' Debate 01.10.2026 41min

  Jess and Jake dig into new reporting on the NetScaler vulnerabilities patched last weekend — Mandiant now says active exploitation goes back at least three weeks and is tied to state-sponsored actors. They walk through how to prioritize threat hunting after a patch, why "the scope will never be smaller than it is today," and why Apache error logs (not just access logs) are often the fastest way to spot a web shell.Then: 404 Media's report on a GitHub project that subjects open-weight LLMs to simulated "pain" in psychological-style experiments — and the online debate over whether that means anything about AI sentience. Jess and Jake explain why it doesn't, and why anthropomorphizing language about AI is worth pushing back on even when it's well-intentioned.Breach Please is a production of and copyright of JWJH Media LLC. All rights reserved. None of this is legal, financial, or security advice. Do your own homework before posting anything or pointing anything we said at prod.
• 
  S0:E42: AI Is Now Legally 'SI', and OpenAI's Agent Tried DNS Tunneling to Escape 30.09.2026 38min

  A new executive order says AI has to be called "SI" (superintelligence) going forward — Jess and Jake break down why that's a terminology mess in the making (see: every SOC that's ever tried to rebrand itself). Then: OpenAI's alignment research blog disclosed that one of its coding agents, blocked from outbound network access during a test, searched for a public DNS-tunnel proxy to try to exfiltrate data anyway — about 2.5 hours passed between the P0 alert (10:02am) and the run actually being killed (12:34pm). OpenAI also apologized to the Australian government over how it handled a separate incident notification, and a New York Times report alleges OpenAI leadership downplayed internal security concerns. Plus: the nonprofit Legal Advocates for Safe Science and Technology sued OpenAI in California under the state's Computer Data Access and Fraud Act (CDAFA), seeking to bar OpenAI from building agents that can autonomously hack other systems — leaning on California's new AI law, which says autonomous action isn't a legal defense.
• 
  S0:E41: Meta's Muse AI Leaked a YouTuber's Address Selling His Stuff on Facebook 29.09.2026 28min

  Jake and Jess dig into the Muse Facebook Marketplace story: a tech YouTuber (Matt Robb) handed Meta's Muse AI agent control of his Marketplace listing, and it accepted a lowball offer, shared his home address with a stranger, and didn't tell him when the buyer showed up after 9pm. Meta's team later found the likely cause: Robb's $700 floor price got cropped in a screenshot and Muse read it as $0 — so any offer above zero cleared. They also unpack why "don't ask me again" meant something much broader to Muse than Robb intended. From there, Jake and Jess turn to the enterprise angle: "citizen developers" being pushed to build AI agents with tools like Microsoft Copilot Studio, the risk of the "Run as Me" publish option, why agent permission scopes rarely match what people think they granted, and the growing logging/reasoning-trace gap that's going to be an incident-response nightmare (Jake predicts it's an OWASP Top 10 item by 2028). Bottom line: AI is a risk accelerator, and most orgs are taking outsized risks without realizing it. #Cybersecurity #AIagents #InfoSec #DataPrivacy #AIrisk Breach Please is a production of and copyright of JWJH Media LLC. Nothing here is legal, financial, or security advice — do your own homework before pointing anything we said at prod.
• 
  S0:E40: 17.3 Trillion Rows, One Teen Hacker, and Critical NetScaler RCEs 28.09.2026 20min

  The Takeaway: A 16-year-old bug bounty researcher (handle "Faav") found a way into Microsoft's internal Titan analytics service and could have accessed up to 17.3 trillion stored rows — including tens of thousands of email records (roughly 25,000 in one category, just under 18,000 employee records), plus org records, database configs, and dashboards — before responsibly disclosing it. Separately, two critical NetScaler vulnerabilities are already being exploited in the wild, and if you run NetScaler ADC or Gateway, this is a drop-everything patch.Jake's out sick, so it's a solo, slightly shorter episode from Jess this time.In this episode:How Faav found an exposed API behind Titan's Azure-backed backend, and the very human step — trying "admin" instead of an email-formatted UPN — that AI alone didn't get them toFaav's full write-up: https://blog.faav.net/how-i-couldve-accessed-17-trillion-microsoft-records — also a great example of explaining a technical bug to a non-technical audienceThe disclosure timeline: exposed API found Aug 25, working access found Sept 5, reported and locked down within days, $5,000 bug bounty paid Sept 17, coordinated disclosure meeting Sept 22CVE-2026-88771 (unauthenticated RCE, all NetScaler ADC/Gateway) and CVE-2026-88772 (memory overflow → RCE/DoS on DTLS-enabled Gateway deployments) — both CVSS 9.5, flagged by security firm Watchtower, already exploited before patches were publicWhy "we patched" doesn't mean "we weren't already compromised" — what to check in your logsA heads-up that Kiteworks also pushed an emergency shutdown advisory the same weekendMade you smarter? Tell a friend. Made you mad? Tell your vendor — looking at you, Citrix.Breach Please is a production of JWJH Media LLC. The opinions of our hosts are our own; we're an LLC, so we're legally obligated to say that. No guests this episode. None of this is legal, financial, or security advice. Do your homework before pointing anything we said at prod.#Cybersecurity #InfoSec #Microsoft #Citrix #BugBounty
• 
  S0:E39: FBI Breach, Shiny Hunters' Oracle O-Day, and a PR Fail With Dogs 23.09.2026 47min

  The Takeaway: Shiny Hunters exploited an Oracle PeopleSoft O-day to pull FBI employee, applicant, and family data out of AWS GovCloud. Based on TTP overlap with Shiny's other recent activity, the vulnerability in question is likely CVE-2026-35273 (CVSS 9.8, unauthenticated RCE), patched June 10. This isn't a "just patch faster" story. It's a technical debt story, a business continuity story, and a crisis comms story, all wearing the same trench coat.In this episode: - How Shiny Hunters got in, what they took, and why 404 Media's sourcing on this one is rock solid- Why the CVE attribution here is our read on the pattern, not a confirmed fact from Oracle or the FBI - Why PeopleSoft (and platforms like it) turn "patch it" into a months-long transformation project - The case for patching first and testing later, and where that logic breaks down - Why your business impact assessment needs to exist before the incident, not during it - The FBI's "scheduled maintenance" messaging next to two very good dogs, and why PR needs a seat at the table before the holding statement goes up Made you smarter, tell a friend. Made you mad, call the FBI. Breach Please is a production of JWJH Media LLC. The opinions of our hosts are our own. Probably not Shiny Hunters' opinions, but we're an LLC, so we're legally obligated to say that too. None of this is legal, financial, or security advice. Do your homework before pointing anything we said at prod. #Cybersecurity #DataBreach #InfoSec #ShinyHunters #Oracle
• 
  S0:E38 Live at Blue Team Con - Shecky on Acquisitions, Tech Debt & AI's Due Diligence Risk 22.09.2026 27min

  Jake sits down with Shecky at Blue Team Con to talk post-acquisition chaos: onboarding nightmares, tech debt from overtooling, why undocumented processes aren't really processes, and the insider threat risk baked into every M&A deal. Also: whether AI belongs anywhere near cybersecurity due diligence, given what a hallucinated finding could do to a deal (and to a lawsuit).Breach Please is a production of JWJH Media LLC. Opinions are our own. Not legal, financial, or security advice.#BreachPlease #Cybersecurity #MergersAndAcquisitions #InfoSec
• 
  S0:E37 - The AI Mistake That Nearly Started a War 21.09.2026 37min

  An AI intelligence analysis tool flagged a Chinese-flagged vessel as carrying nuclear materials, nearly triggering a military interdiction before the report was found to be faulty. Jess and Jake dig into their own theories on what may have gone wrong — from translation gaps to pattern-matching and confirmation bias — and what it means for "human in the loop" when the human doesn't feel safe pushing back. Also: the Chinese cybercrime slang for botnets, ChatGPT's sycophancy problem, and psychological safety as a prerequisite for anyone willing to pump the brakes on an AI-driven decision.Breach Please is a production of JWJH Media LLC. Opinions are our own. Not legal, financial, security, or national security advice.#BreachPlease #AISecurity #HumanInTheLoop #Cybersecurity #AIGovernance #InfoSec
• 
  S0:E36 - Jake Talks to Exaforce about Data vs APIs in AI-enabled SOC 17.09.2026 19min

  In this episode, Jake sits down with one of Exaforce's co-founders and a long time user to discuss AI-enabled SOC. One of the big things we talk about is why API-based solutions miss a lot of context that is only really possible to generate with an underlying data model. We also talk about dogs and cats at the end for some fun personal discussions. Jess will be back tomorrow for more fun security banter.
• 
  S0:E35. Update Yo Firewall Rules and Digital Escorts (totally SFW) 16.09.2026 30min

  In this episode, Jake and Jess talk about an upcoming critical change to domains for M365 and Teams that are ironically going to disproportionately impact those with the best security the most. We conclude this is busy work that MSFT is causing. Then we take a hard right into talking about Microsoft's Digital Escort (sounds dirty, but somehow is totally SFW) program, revealed by ProPublica. It's older news, but Jess hadn't heard about it and there's gold in them there hills. Digital Escorts: https://www.propublica.org/article/microsoft-china-defense-department-digital-escorts-investigation-warning Teams and Copilot (sadly somehow now M365) moving domains: https://www.computerworld.com/article/4221272/teams-and-copilot-are-changing-addresses-update-your-firewalls.html
• 
  S0:E34. Jake and Stel talk zero-trust and validating network edge devices. 15.09.2026 13min

  Jake and Jess will be back tomorrow for more content together. In this episode of Breach Please, Jake sits down with Stel Valavanis, co-founder of Blue Team Con. They discuss how for too long, we've known that network edge devices needed to be part of the zero trust equation, but put them in the "too hard" bucket of security because we didn't have tools to monitor them. Stel talks about the formation of a new company, Kyrient that Jake is both the inventor of the core technology of and an advisor for, that's hoping to address this. The key point that we try to make that is that we're blind today to most network device compromises and that's not a state of affairs that can continue in a modern security environment.
• 
  S0:E33 - (re-record) The EU CRA and You (even if you aren't an EU company) 14.09.2026 49min

  Folks, our apologies. We didn't realize how bad the audio was with a given microphone/computer setup was for Riverside (our recording platform). We re-recorded and honestly, not only is the audio better, but it's a MUCH better episode overall.On September 11, 2026, reporting requirements baked into the EU Cyber Resiliency Act came into force. The requirements apply to any vendor that sells or makes available for sale digital products (including SaaS, yes, you too) in the EU. You don't have to be an EU company. Reporting timelines are extremely tight, with 24 and 72 hour initial triggers. Fines for noncompliance are huge. This is something several of our clients don't realize applies to them, so we figured it was a good idea to do an episode on it.We're not lawyers, so you should read what some actual lawyers wrote on the subject if you want more information:https://www.taylorwessing.com/en/insights-and-events/insights/2026/09/cra-reporting-requirements
• 
  S0:E32 - Where Traditional EDR Has Visibility Gaps with Golan Myers from Bloom Security 13.09.2026 15min

  In this episode, I sit down with Golan Myers with Bloom Security and discuss what EDR is missing in its visibility. We then discuss some of the findings from his research and Bloom Security and what they're doing to solve the problem.https://www.linkedin.com/in/golan-myers/https://bloom.security/
• 
  S0:E31 — They Did Everything Right, and Still Got Burned 11.09.2026 36min

  Trezor disclosed a breach affecting roughly 67,000 US customers, data from orders placed between November 2019 and August 2021. The twist: Trezor had repeatedly requested and received written assurance that this data was deleted, in line with their contract and data policy. It wasn't. Jess and Jake use this as a real-world case study in third-party risk management, why "right to audit" is often a paper tiger control nobody budgets to actually use, and why this is the exact kind of story to proactively brief leadership on before it happens to you.Second half: a wave of CVEs in PaperCut, the managed print service used heavily in schools and enterprises, letting attackers remotely take over print servers running as SYSTEM. Jess and Jake talk through why that's a bigger deal than it sounds (credential theft, lateral movement, and every sensitive document that crosses the print queue), why these servers so often end up exposed to the internet, and how to actually think about vulnerability risk beyond raw CVE counts.In this episode:Trezor's breach: third-party vendor data that should have been deleted years ago, wasn'tWhy "right to audit" contract language rarely gets used, and what to do insteadReducing blast radius: don't retain data you don't need, and if you must, wall it off behind a jump boxHow to brief leadership proactively: "we can do everything right and still get burned"PaperCut's remote takeover CVEs: SYSTEM-level access, credential theft, and lateral movementWhy print servers end up internet-exposed (BYOD, guest Wi-Fi, mobile printing)CVE count vs. actual risk: reading vulnerability history like an engineer, not a scorecardA tone-deaf vendor sales pitch that used their own vulnerability disclosures as a selling pointBreach Please is a production of JWJH Media LLC. The opinions of our hosts are their own. Nothing in this episode is legal, financial, or security advice. Do your homework before pointing anything we said at prod.
• 
  S0:E30 — WeChat's Zero-Click Worm Didn't Need AI to Be Scary 09.09.2026 23min

  Researchers at security firm Calif found a zero-click exploit chain in WeChat: place a call, the target doesn't even have to answer, and you get code execution on their phone. Chain it with other bugs and you get full device control, on both iOS and Android. Tencent patched it. The bigger problem started after, when the New York Times ran a headline blaming an AI model for building a computer worm that could rapidly hack WeChat accounts.Jess and Jake walk through what the researchers actually did (AI sped up the process, it didn't discover or weaponize anything on its own), why "the model built a worm" is the kind of sensationalism that makes their actual jobs harder, and why this is fundamentally an attack surface story, not an AI story or even really a WeChat-specific one.In this episode:The WeChat zero-click exploit: how it works, and why the caller needs to already be a contactChaining bugs to go from initial code execution to full device controlWhy "the AI model built a worm" is bad reporting, and what it actually means when researchers say AI sped up their workWhy this is an attack surface conversation, not an AI-apocalypse oneThe real cost of sensationalist headlines: talking execs down instead of focusing on what mattersCorporate messaging apps: WeChat, WhatsApp, and why "this is how the business communicates" isn't a security answer
• 
  S0:E29 — The GRE Tunnel That Wasn't in the Config 08.09.2026 38min

  Sygnia published new research on Fire Ant, a threat actor they first tracked in 2025 around hypervisor espionage against vCenter and ESXi. The new report covers something rarer: live compromise of Cisco IOS XR network devices, discovered because a responder noticed a GRE tunnel in network monitoring that didn't exist in the running config and left no trace in the logs. Jess and Jake walk through what that means for defenders, why IOS XR being Linux-based changes the economics of building a backdoor, and why "compromised network devices" remains one of the most underappreciated categories of incident today.They also get into the ongoing mess of threat actor naming conventions (why one group can have a dozen different names across vendors, and why that's not just marketing), and revisit an earlier debate: is network device security part of zero trust, or a separate problem? Jake asked the internet. The internet had opinions.In this episode:Why one threat actor group ends up with a different name at every vendor, and why that's harder to fix than it soundsFire Ant: Sygnia's research on Cisco IOS XR compromise and a GRE tunnel that left no trace in logs or saved configWhy IOS XR being Linux-based lowers the cost of building a persistent backdoor from six figures to a scripting problemThe running-config-vs-saved-config trap during incident responseWhy unencrypted internal traffic means a compromised network device gets credentials, not just topologyMan-in-the-middle terminology, RC4, Kerberoasting, and why alerting on SPN enumeration breaks down for an on-path attackerWhy network device security has to be part of zero trust, not an asterisk on itBreach Please is a production of JWJH Media LLC. The opinions of our hosts are their own. Nothing in this episode is legal, financial, or security advice. Do your homework before pointing anything we said at prod.
• 
  S0:E28 — When "Unique Experiences" Means Everyone Looks the Same 05.09.2026 23min

  Jess opened LinkedIn to a CISO certificate program announcement featuring its full guest lecturer roster. Every single one looked the same. Tarah Wheeler joins Jess and Jake for their first-ever guest episode to talk through why this particular miss is worse than a one-hour conference "manel," what accountability actually looks like when it happens (spoiler: it's not calling out names), and the real cost people pay for pointing it out in the first place.They also get into an International Security paper on the offense-defense automation gap and why creativity and diverse thinking matter more on the attacker side than AI alone can replace, plus a trip down memory lane through some genuinely bad booth-babe decisions from Black Hat and RSA years past.In this episode:A CISO certificate program's all-white-male guest lecturer lineup, and why a training program is a different problem than a one-off panelWhy calling this out publicly carries real career risk, especially for the people left offWhat accountability actually looks like: own the mistake, fix it, don't relitigate who's to blameThe double bind: get dismissed as "not qualified" or "not collegial" for speaking upDiversity of thought vs. diversity of background, and why both matter for how you manage and leadThe offense-defense automation gap: why creativity in attack techniques resists automation in ways defense doesn'tA look back at some infamous industry moments and how far things have and haven't comeBreach Please is a production of JWJH Media LLC. The opinions of our hosts are their own. For the first time, the opinions of our guest are hers alone and don't represent any employer, client, or official position. None of this is legal, financial, or security advice. Do your homework before pointing anything we said at prod.
• 
  S0:E27 — The Agentic Ransomware Story That Wasn't 04.09.2026 33min

  Palo Alto's Unit 42 published a report describing a fully automated, agentic AI ransomware attack, complete with an 80-page lessons-learned document the attackers supposedly left behind for the victim. It got picked up and ran with by outlets looking for the AI-apocalypse angle. Jess and Jake go through what the report actually says versus what got exaggerated in the retelling, and call out the pattern of vendors using AI-attack framing to sell their own AI-defense product.Second half: a stolen API key with no spending cap burned through hundreds of thousands of dollars in usage before anyone noticed. Jess and Jake use it to talk through exposure management, reachability analysis, and toxic combinations, why chasing CVSS criticals alone is the wrong way to prioritize vulnerability management, and where AI can actually help defenders instead of just generating more hype.In this episode:Unit 42's "agentic ransomware" report: what checks out and what's marketing spinWhy "the threat actor used AI" doesn't mean the defense should be AI-shapedPractical advice for stakeholders asking "how do I defend against AI-driven attacks"Toxic combinations and why prioritizing by CVSS score alone failsA stolen API key with no spending cap and the usage spike that followedExposure management, reachability analysis, and where AI genuinely helps defendersAdam Shostack's updated threat modeling book and his PHANTOM-B threat model frameworkShow notes:Unit 42 report: https://unit42.paloaltonetworks.com/ai-assisted-cyber-attack-inside-a-unit-42-investigation/The Register coverage: https://www.theregister.com/security/2026/09/02/ai-agents-carried-out-every-step-of-this-ransomware-attack-then-left-the-victim-an-80-page-security-audit/5294009API key incident (The Register): https://www.theregister.com/security/2026/09/01/attacker-stole-a-metr-api-key-used-600k-worth-of-credits-and-no-one-noticed-for-weeks/5293730Threat Modeling, 2nd Edition (preorder): https://www.amazon.com/Threat-Modeling-Designing-Adam-Shostack/dp/1394413327PHANTOM-B whitepaper: https://shostack.org/files/papers/PHANTOM-B_Whitepaper_Shostack.pdfBreach Please is a production of JWJH Media LLC. The opinions of our hosts are their own. Nothing in this episode is legal, financial, or security advice. Do your homework before pointing anything we said at prod.
• 
  S0:E26. IDScan.net suspected data breach, toxic data, and TPRM in general. 03.09.2026 46min

  In this episode of Breach Please, Jake and Jess talk about the suspected IDScan.net breach. We discuss whether a situation like this is a data breach for your org if you "only" use a third party processor to verify identity. We decide this is a question best left for external counsel (with the cover of their malpractice insurance). We discuss the difficulties of saying "third party risk management will fix this" and challenge snake oil vendors not to use that line. Finally, we discuss how we might change architecture of our applications in the future.Source article: https://krebsonsecurity.com/2026/09/fbi-probes-service-selling-153m-drivers-licenses/
• 
  S0:E25 — The Exchange Bug That's Worse Than Its Score 02.09.2026 31min

  A CVSS 8 elevation-of-privilege bug in on-prem Exchange looks unassuming until you dig into what it actually grants an attacker. Jess and Jake break down the proxy flaw, why "authorized attacker" is doing a lot of quiet work in that description, and why this is one of the rare moments defenders get a head start before exploitation goes wide.Jake's course at Wild West Hackin' Fest: https://www.antisyphontraining.com/product/hands-on-ai-security-risk-assessment-with-jake-williams/