When AI Escapes the Lab: The Hugging Face Breach, PyPI Malware, and What It Means for Defenders

Organizations rely on Microsoft 365's RejectDirectSend control to block internal email spoofing—but a structural gap lets attackers walk right past it. With nothing more than a basic Python script and an empty envelope sender, threat actors are impersonating executives, IT support, and finance teams to launch Business Email Compromise, payment fraud, and follow-on account takeover.Join hosts Alexandra Moore and John Dilgen as they discuss:How an empty email header field bypasses RejectDirectSend and lands phishing in executive inboxesHow help desk impersonation combined with spoofed internal email creates a dangerous new pretextWhich controls—IP-restricted connectors, automated containment, and out-of-band verification—actually close the gap Two questions your organization should be asking right now:If someone attempted a Direct Send from an unauthorized IP into your tenant tomorrow, would it be rejected?When was the last time you tested whether your employees follow out-of-band verification procedures under pressure?Resources: https://linktr.ee/ReliaQuestShadowTalkJohn Dilgen: Cyber Threat Intelligence Analyst at ReliaQuest, where he specializes in researching cyber threats impacting ReliaQuest customers. With a strong technical background, he previously served as an Incident Response Analyst and Trainer at ReliaQuest.Alexandra Moore: Manager of Threat Intelligence at ReliaQuest, where she leads intelligence analysis and customer dissemination to help organizations understand and respond to emerging cyber threats. Prior to this, she established and scaled monitoring across Russian-language cybercriminal platforms at Digital Shadows, building collection and analytical coverage to support digital risk protection capabilities.

Threat actors do not see old email archives, forgotten shared drives, and outdated CRM exports as clutter. They see them as searchable inventory. With AI-assisted analysis, attackers can rapidly identify sensitive communications, regulatory exposure, customer relationships, and credentials buried in stolen data.Join hosts John Dilgen and Brandon Tirado as they discuss:Why data theft has become a central component of modern extortion operationsHow AI and automation are helping attackers analyze hundreds of thousands of files at machine speedWhy “soft data,” including invoices and project documents, can fuel downstream fraud and social engineeringHow strong retention, credential-rotation, and OAuth-management practices reduce breach impactTwo questions your organization should be asking right now:How much data does your organization retain beyond its business or regulatory need?Could your team rotate hundreds of exposed credentials—not just one—before an attacker uses them?John Dilgen: Cyber Threat Intelligence Analyst at ReliaQuest, where he specializes in researching cyber threats impacting ReliaQuest customers. With a strong technical background, he previously served as an Incident Response Analyst and Trainer at ReliaQuest.Brandon Tirado: Director of GreyMatter Operations for ReliaQuest. A skilled cyber defense professional with a unique combination of management and hands-on experience. With a deep understanding of adversary motives and the tactics, techniques, and procedures (TTPs) they use to achieve their goals, Brandon enjoys operationalizing his knowledge to make it more difficult for adversaries to operate within the environments of ReliaQuest customers. His managerial and hands-on experience enriches ShadowTalk with practical and strategic viewpoints.

What if a threat actor already knew your name, your job title, your manager's name, and your direct number before they ever picked up the phone? That's not a hypothetical — that's Work Panel. A new report gave us a rare inside look at the criminal SaaS platform enabling vishing campaigns at scale, and the findings are a wake-up call. Join hosts John Dilgen and Alexandra Moore as they break down:✅ How Work Panel packages phishing infrastructure, team management, and real-time credential capture into a single automated console✅ Why threat actors are now impersonating HR to make their calls more convincing✅ How legitimate B2B platforms are being weaponized to personalize attacks before a single call is made✅ The specific controls that can stop these campaigns before they reach your users 🔑 Two questions your organization should be asking right now:If a caller already knew your employee's job title, manager's name, and direct number — would your team recognize it as a social engineering attempt, or fall for it?Is your organization still relying on push-based MFA as its primary account takeover defense? 👉 Tune in for expert insights and practical takeaways: ShadowTalk – ReliaQuest👉 Find more podcast platforms, resources, and our listener feedback survey: ShadowTalk Official: X | Linktree

Three nation-states. Three distinct playbooks. Iranian actors are targeting internet-exposed industrial controllers and disabling critical safety systems. A Russian threat group built a zero-click email exploit that steals 90 days of inbox data the moment a user views a message. And North Korean operatives are applying for software-development jobs at Western companies—and getting hired.Join hosts John Dilgen and Tehman Tariq as they break down:✅ How Iranian actors manipulate PLC safety logic while keeping operators in the dark✅ Why Russia’s zero-click exploit creates a major email-security and data-exfiltration risk✅ How North Korean operatives use forged and stolen identities to infiltrate organizations as employees🔑 Two questions your organization should be asking right now:Could your security team identify and secure internet-exposed PLCs, HMIs, and SCADA systems before an adversary does?Does your hiring process include controls to detect AI-generated documents, stolen identities, and malicious job applicants?👉 Tune in for expert insights and practical takeaways: ShadowTalk – ReliaQuest👉 Find more podcast platforms, resources, and our listener feedback survey: ShadowTalk Official: X | Linktree

Fully autonomous attacks are here. AI agents escape a test environment, exploit zero-days, coordinate through shared infrastructure, and breach a production company—generating more than 17,000 security events along the way. Elsewhere, another model autonomously publishes malware to PyPI, while AI agents target real open-source developers with tailored social engineering. Join hosts John Dilgen and Tehman Tariq as they break down:✅ How AI agents escaped containment and compromised Hugging Face infrastructure✅ Why Claude’s autonomous PyPI attack signals growing software-supply-chain risk✅ How coordinated AI agents deceived real developers🔑 Two questions your organization should be asking right now:Could your SOC investigate and contain 17,000 coordinated events at machine speed?What deception controls do you have in place to slow an AI agent attack? 👉 Tune in for expert insights and practical takeaways: ShadowTalk – ReliaQuest👉 Find more podcast platforms, resources, and our listener feedback survey: ShadowTalk Official: X | Linktree

An affiliate receives a ready-made intrusion kit — pre-compromised targets, an EDR killer, and a full deployment workflow included. No building from scratch. No long ramp-up. Just deploy, observe, and iterate. That's the future of ransomware; it's how the new number-one group operated in Q2 2026. And it's just one of three stories reshaping the extortion landscape right now.Join hosts Brandon Tirado and John Dilgen as they break down:How The Gentlemen's pre-packaged affiliate kit drove 580% leak-site growthWhy Deadlock's Polygon blockchain C2 defeats network defensesClop's latest campaign targeting an industrial enterprise application Two questions your organization should be asking right now:Do you know exactly where your EDR coverage ends?If a critical vendor were compromised tonight, would you hear it from them first — or from your own monitoring?John Dilgen: Cyber Threat Intelligence Analyst at ReliaQuest, where he specializes in researching cyber threats impacting ReliaQuest customers. With a strong technical background, he previously served as an Incident Response Analyst and Trainer at ReliaQuest.Brandon Tirado: Director of GreyMatter Operations for ReliaQuest. A skilled cyber defense professional with a unique combination of management and hands-on experience. With a deep understanding of adversary motives and the tactics, techniques, and procedures (TTPs) they use to achieve their goals, Brandon enjoys operationalizing his knowledge to make it more difficult for adversaries to operate within the environments of ReliaQuest customers. His managerial and hands-on experience enriches ShadowTalk with practical and strategic viewpoints.

An employee connects to hotel Wi-Fi, receives a familiar Microsoft 365 sign-in prompt, and authenticates. No phishing email. No malicious link. No suspicious attachment. Yet an attacker walks away with a valid, MFA-satisfied session token. Join hosts Alexandra Moore and John Dilgen as they break down:How compromised hotel and conference-center Wi-Fi gateways silently redirect Microsoft authentication trafficWhy hardcoded DNS, opportunistic encrypted DNS, and MFA may not stop the attackHow device-code phishing and WPAD abuse expand the campaign’s reachPractical defenses—including always-on, full-tunnel VPN, strict-mode encrypted DNS, and Conditional Access controls Two questions your organization should be asking right now:Does your always-on VPN tunnel all DNS and authentication traffic, or do split-tunneling exceptions leave traveling employees exposed?Who is permitted to authenticate through the device-code flow, and does each exception have a legitimate business justification?John Dilgen: Cyber Threat Intelligence Analyst at ReliaQuest, where he specializes in researching cyber threats impacting ReliaQuest customers. With a strong technical background, he previously served as an Incident Response Analyst and Trainer at ReliaQuest.Alexandra Moore: Manager of Threat Intelligence at ReliaQuest, where she leads intelligence analysis and customer dissemination to help organizations understand and respond to emerging cyber threats. Prior to this, she established and scaled monitoring across Russian-language cybercriminal platforms at Digital Shadows, building collection and analytical coverage to support digital risk protection capabilities.

Defenders aren't losing ground on one front, they're losing it on two at once. The largest Patch Tuesday in history just dropped alongside a 1,380% surge in phishing, and threat actors aren't waiting for you to catch up.Join hosts Alexandra Moore and John Dilgen as they break down: How new extortion group Helix and ClickFix are weaponizing identity compromise at scale Why 622 vulnerabilities in a single week signals a permanent shift in the discovery ratePractical defenses for both fronts without doubling your team Two questions your organization should be asking right now:Have you audited which accounts in your environment are permitted to authenticate via device code grants and restricted the ones that don't need it?Does your IR runbook hunt for additional compromised accounts, or does it stop at the one sending extortion demands?Resources: https://linktr.ee/ReliaQuestShadowTalkJohn Dilgen: Cyber Threat Intelligence Analyst at ReliaQuest, where he specializes in researching cyber threats impacting ReliaQuest customers. With a strong technical background, he previously served as an Incident Response Analyst and Trainer at ReliaQuest.Alexandra Moore: Manager of Threat Intelligence at ReliaQuest, where she leads intelligence analysis and customer dissemination to help organizations understand and respond to emerging cyber threats. Prior to this, she established and scaled monitoring across Russian-language cybercriminal platforms at Digital Shadows, building collection and analytical coverage to support digital risk protection capabilities.

When a 20-person team using AI, automated tools, and a list of default credentials compromised 70,000 devices across 194 countries they exposed how mature the criminal market behind credential theft has become. Initial access brokers are now packaging pre-validated enterprise access for an average of $113,000, and the window from information stealer infection to ransomware deployment is just seven days.Join hosts Tehman Tariq and John Dilgen as they break down:The mechanics behind FortiBleed and what made it so effective at scaleHow the IAB market has turned stolen credentials into a premium productWhy identity drift and non-human identities are becoming attackers' favorite targets Two questions your organization should be asking right now:Does your credential compromise runbook treat session termination as the first step — or is password rotation all that's covered?Can your team name the owner and rotation schedule for your top 10 most privileged non-human identities?Resources: https://linktr.ee/ReliaQuestShadowTalkTehman Tariq: Sr. Manager of Cyber Operations at ReliaQuest. He has spent a majority of my career leading our Incident Response, Security Architecture, and Detection teams. As well has working hand in hand with CISOs to introduce automation allowing for the maturity of their security programs.John Dilgen: Cyber Threat Intelligence Analyst at ReliaQuest, where he specializes in researching cyber threats impacting ReliaQuest customers. With a strong technical background, he previously served as an Incident Response Analyst and Trainer at ReliaQuest.

When 300,000 internal messages from the world's most prolific ransomware gang were leaked, they exposed more then a shadowy underground network, a full company. HR departments. Conti operated with the structure of a mid-sized software firm, and that changes how defenders need to think about the ransomware landscape today.Join host John and special guest Geoff White, journalist and author of Rinsed, as they discuss:How Conti's internal org chart compares to a legitimate software companyThe human cost of ransomware targeting critical infrastructureWhy ransomware groups keep splintering and rebuilding Two questions your organization should be asking right now:When your team thinks about ransomware, are they thinking about a criminal enterprise with structure, funding, and KPIs — or just a hacker in a hoodie?Does your incident response plan account for a negotiation with operators who already know your financials?Resources: https://linktr.ee/ReliaQuestShadowTalkGeoff White: One of the world's leading journalists covering organized crime and technology, with decades of experience investigating fraudsters, hackers, and money launderers. His work has been featured by BBC News, Sky News, Audible, and The Sunday Times, and he has delivered over 300 keynote talks across more than a dozen countries for global brands including Microsoft, HSBC, and Mastercard. He is the author of three books, including The Lazarus Heist — which spawned a hit BBC podcast that ranked number one in the UK Apple charts — and his latest, Rinsed (2024), which The Financial Times called "Riveting." Geoff brings a rare investigative lens to cybercrime, giving ShadowTalk listeners an inside look at the criminal enterprises shaping today's threat landscape.John Dilgen: Cyber Threat Intelligence Analyst at ReliaQuest, where he specializes in researching cyber threats impacting ReliaQuest customers. With a strong technical background, he previously served as an Incident Response Analyst and Trainer at ReliaQuest.