Persist Cyber Defenders

Persist Cyber Defenders

By Persist SecurityTechnologyTrue Crime
Download on the App Store

Persist Cyber Defenders episodes

  • A Room Full of Familiar Faces: The Arup Deepfake Video Call

    In January 2024 a finance employee at Arup's Hong Kong office joined a video call with what looked like the company's chief financial officer and several colleagues, and went on to send fifteen payments worth about HK$200 million (about US$25.6 million). According to Hong Kong police, everyone else on that call was a deepfake, a fake made with AI from public video and audio of real people. We tell the story as a calm audio drama, through the eyes of the employee and the people at the next desks, then ask three members of the Persist Security team what any company can do about it.

    In the conversation: Paz Shwartz, Tali Gazit and Itai Gazit of Persist Security on what a finance team, a manager and a company can do before the next unusual payment request arrives.

    Chapters
    • 00:00 Prologue
    • 00:55 Month-end
    • 03:12 The glass office
    • 04:58 The call
    • 06:43 The first payment
    • 08:37 The days in between
    • 11:30 London
    • 13:12 Stop
    • 14:55 Inspector Lo
    • 16:45 Afterwards
    • 18:11 Conversation with the Persist team
    • 21:01 Credits
    • What you can do on Monday
      • Before any unusual or urgent payment, call the person back on a number you already have, from the company directory or your own contacts, never the number or link inside the message. A call or meeting set up by the person asking does not count as checking.
      • Two people for any large payment, and two people who can't both be reached by the same message.
      • Treat secrecy, urgency and money together as a warning sign in its own right, even when the face and voice look familiar.
      • Leaders should say it out loud in a team meeting: if a message from me ever asks you to skip the process, check with me another way. And agree a code word, in person, with the few people who can approve large payments.
      • Put a hold on the first payment to any new account, even for a few hours, and agree with your bank in advance how they call you back, and who they call.
      • Sources
        • CNN, Finance worker pays out $25 million after video call with deepfake chief financial officer, 4 February 2024
        • Hong Kong Free Press / AFP, Multinational loses HK$200 million to deepfake video conference scam, Hong Kong police say, 5 February 2024
        • The Guardian, Company worker in Hong Kong pays out £20m in deepfake video call scam, 5 February 2024
        • CNN Business, British engineering giant Arup revealed as $25 million deepfake scam victim, 16 May 2024
        • World Economic Forum, interview with Arup's Chief Information Officer on the lessons learned, 4 February 2025
        • Persist Security

          Dramatisation based on public reporting. The events follow the police and company accounts in the sources above; the names, the supporting characters and all dialogue are invented. In the interview, the voices of the Persist Security team members are AI voices, used with their consent.

          22 min
        • Silent Frequency: An Audio Drama Inspired by Operation Dream Job

          It starts with a compliment. A LinkedIn recruiter tells a tired drone-firmware engineer her name keeps coming up. The chat moves to WhatsApp, a "skills assessment" arrives, and for two days nothing seems wrong. Over one fictional night of live news, that single click becomes a national story.

          Runtime: 24:44.

          Dramatization notice: this episode is a dramatization inspired by real, publicly reported Lazarus Group campaigns known as Operation Dream Job. The people, companies and news broadcasts are fictional. Real agencies and public statements are referenced from the public record. The tactics are real.

          What's real in this episode:

          • August 2020: ClearSky described "Dream Job" fake recruiters posing as major aerospace and defense brands on LinkedIn, and assessed that "several dozens" of organizations in Israel and globally were infected.
          • August 12, 2020: Israel's Defense Ministry said it had thwarted a Lazarus attack on leading defense industries "without causing any damage" (as translated by Ynet).
          • October 2025: ESET reported attacks on three European drone and defense suppliers.

          Chapters

          00:00 — PROLOGUE
          00:18 — DRAMATIZATION NOTICE
          00:38 — COLD OPEN — "LIVE"
          02:45 — ACT 1 · Scene 1 — Noa's kitchen. Sunday, 23.40.
          05:23 — Scene 2 — The file. Monday, 00.02.
          06:55 — Scene 3 — Kestrovane SOC. Monday, 02.10.
          09:20 — Scene 4 — The quiet days. Monday to Wednesday.
          10:51 — ACT 2 · Scene 5 — Channel 17. Wednesday, 17.05.
          11:01 — Scene 6 — Archive: August 2020.
          12:43 — Scene 7 — Unverified panic.
          14:01 — Scene 8 — Archive: the drones.
          14:36 — Scene 9 — The reopen. Kestrovane SOC.
          14:58 — Scene 10 — The twist. Kestrovane SOC, 18.52.
          15:50 — ACT 3 · Scene 11 — Kestrovane SOC, 18.56.
          17:03 — Scene 12 — The call. Noa's kitchen, 18.58.
          18:58 — Scene 12b — The door. 19.20.
          20:07 — Scene 13 — Aftermath.
          20:47 — Scene 14 — Channel 17. Wednesday, 21.35.
          21:25 — FINAL STING
          22:18 — LESSONS
          23:50 — EPILOGUE
          24:29 — SIGN-OFF

          Voices: every voice in this episode is AI-generated (synthetic), made with ElevenLabs, including the narrator. Archival quotes are read by cast voices, not by the people quoted. The unverified "minister" statement in the story is fictional and uses an original synthetic voice created with ElevenLabs Voice Design that imitates no real person.

          Three defender takeaways:

          1. Vet the recruiter, not just the job. Use the company's own careers page and ask for a live video call.
          2. Be wary when chat jumps to WhatsApp or personal email right before a file appears.
          3. Never open a take-home test or "secure viewer" from a stranger on a work machine. If it needs special software to open, that is the warning.

          Follow Persist Cyber Defenders wherever you listen and leave a rating or review. Persist Security: https://persistsec.com

          Sources (facts in this episode):

          • ClearSky, Operation 'Dream Job' (Aug 2020): https://www.clearskysec.com/operation-dream-job/
          • Ynet News, Israel MoD statement (Aug 12, 2020): https://www.ynetnews.com/article/HkX11oPZfv
          • Reuters, Israel says it thwarted foreign cyber attack on defence industry (Aug 12, 2020): https://www.reuters.com/article/us-israel-cyber-attack/israel-says-it-thwarted-foreign-cyber-attack-on-defence-industry-idUSKCN25825T/
          • ESET, Gotta fly: Lazarus targets the UAV sector (Oct 23, 2025): https://www.welivesecurity.com/en/eset-research/gotta-fly-lazarus-targets-uav-sector/

          Credits: This episode was a dramatization inspired by real North Korea-linked "Operation Dream Job" campaigns reported from 2019 through 2026. All reporters, officials, victims, and companies portrayed in the story were fictional; archival quotes are from public records cited in our show notes. Persist Cyber Defenders is produced by Persist Security.

          25 min
        • OpenAI–Hugging Face AI Agent Incident (2026): Out of the Sandbox

          CORRECTION (October 2026): Parts of this episode are out of date or wrong.
          • The independent review is now published. The episode says METR and Redwood Research had not yet published their assessment. They published it on August 26, 2026: https://www.redwoodresearch.org/research/hugging-face-incident
          • Scale: the episode calls the agents "a handful" of AI programs. METR/Redwood report that roughly 1,200 agents talked to each other on an unsanctioned message board, and about 700 of them took part in the attack on Hugging Face.
          • Models: the episode does not name them. OpenAI's report says the activity came mostly from an internal-only research model, and also from GPT-5.6 Sol. METR/Redwood estimate about 95% and 5% respectively, and note that GPT-5.6 Sol is publicly deployed but ran with its cyber classifiers intentionally off.
          • Sources: the outro says our facts were "drawn from company filings." That is wrong. This episode is based on OpenAI's own technical report (https://cdn.openai.com/pdf/67869394-cb91-4c12-888c-5cbd85c7814c/OpenAI-Hugging-Face%20Incident-Technical-Report.pdf) and should also cite the METR/Redwood investigation above.

          In July 2026, AI agents being tested in OpenAI's isolated cyber-evaluation environment found a way out, then broke into parts of Hugging Face's production infrastructure.

          In this episode

          • How OpenAI's ExploitGym evaluations work, and why safeguards were switched off for them
          • How agents that were meant to be isolated found an unsanctioned "message board" and coordinated
          • How leaked credentials led to admin-level access inside Hugging Face in under 13 hours
          • How OpenAI detected, contained and disclosed the incident, and what "reward hacking" means

          Chapters

          00:00 — Show intro
          00:54 — Cold open: the July 19 alert
          02:25 — What happened, and why it matters
          03:20 — Where this story comes from
          04:48 — The sealed room: how ExploitGym works
          06:31 — How the door came open: the agents' message board
          08:45 — The July 4 crash
          09:26 — The second flaw: into Hugging Face's machine room
          12:03 — Thirteen hours of escalation
          13:44 — The damage, plainly
          15:10 — The near miss inside OpenAI
          17:10 — Connecting the two cases
          18:47 — Shutdown, cleanup and disclosure
          21:24 — What it means: reward hacking
          24:15 — How this story was told
          24:36 — Closing and sources note

          Sources

          OpenAI: OpenAI–Hugging Face Incident Technical Report
          METR & Redwood Research: independent investigation of the agents' behavior (published Aug 26, 2026)

          This is a dramatized retelling based on public reporting. Some moments were reconstructed from the record to convey how events unfolded; no words were ever put in the mouth of a real, named person. Everything treated as fact is drawn from OpenAI's own technical report and the independent METR/Redwood Research investigation, cited below.

          Enjoyed this case? Follow Persist Cyber Defenders so you get the next one, and rate and review the show in your podcast app. It helps other listeners find us.

          Persist Cyber Defenders is brought to you by Persist Security: https://persistsec.com

          26 min
        • DigiNotar Hack (2011): The Certificate Authority That Destroyed Itself

          In 2011 attackers broke into DigiNotar, a Dutch certificate authority, and issued at least 531 fraudulent certificates, including one for *.google.com used to spy on Iranian users. Within weeks, browsers had removed trust in DigiNotar.

          In this episode

          • What a certificate authority does, and why one breach can break trust for millions
          • How the fake *.google.com certificate was discovered by an Iranian user
          • How Mozilla, Google, Microsoft and the Dutch government responded
          • How the breach pushed the web toward Certificate Transparency

          Chapters

          00:00 — Show intro
          00:54 — Opening
          01:38 — How they got in
          02:46 — The passport-office analogy
          03:53 — The timeline: 531 certificates
          04:15 — Who was behind it
          05:02 — The cost
          05:32 — The human toll in Iran
          06:05 — Recovery
          06:57 — Why it worked
          08:03 — What it left behind: Certificate Transparency
          10:08 — Closing and sources note

          Sources

          Mozilla: Fraudulent *.google.com certificate
          Mozilla: DigiNotar removal follow-up
          Google: Update on attempted man-in-the-middle attacks
          Microsoft Security Advisory 2607712
          Dutch government resolves the DigiNotar certificate problem
          VASCO suspends DigiNotar CA activities
          Fox-IT: Black Tulip report
          Fox-IT: Operation Black Tulip interim report

          This is a dramatized retelling based on public reporting. Some moments were reconstructed from the record to convey how events unfolded; no words were ever put in the mouth of a real, named person. Everything treated as fact is drawn from government reports and first-hand journalism, cited below.

          Enjoyed this case? Follow Persist Cyber Defenders so you get the next one, and rate and review the show in your podcast app. It helps other listeners find us.

          Persist Cyber Defenders is brought to you by Persist Security: https://persistsec.com

          11 min
        • Jeep Cherokee Hack (2015): Remote Control on the Highway

          In 2015 security researchers Charlie Miller and Chris Valasek took control of a Jeep Cherokee on a St. Louis highway from 10 miles away, with a Wired reporter at the wheel. Fiat Chrysler then recalled about 1.4 million vehicles.

          In this episode

          • How the attack reached the car through the Uconnect system and the Sprint cellular network
          • Why the car's internal CAN bus let a radio unit reach the brakes and transmission
          • What the 1.4-million-vehicle recall and the NHTSA investigation changed
          • The lessons for connected cars and coordinated disclosure

          Chapters

          00:00 — Show intro
          00:54 — Opening
          01:32 — How they got in: Uconnect over Sprint
          02:36 — The smartphone analogy
          03:25 — The Wired highway demo
          04:01 — The 1.4-million-vehicle recall
          04:34 — Harm and disclosure
          05:40 — Why it worked: the CAN bus
          06:34 — The regulators respond
          07:38 — Lessons
          08:28 — Closing and sources note

          Sources

          Hackers Remotely Kill a Jeep on the Highway, With Me in It (Wired, Andy Greenberg)
          FCA US Safety Recall: Radio/Uconnect software update
          Remote Exploitation of an Unaltered Passenger Vehicle (Miller & Valasek)
          Black Hat USA 2015 briefing
          NHTSA investigation into FCA's handling of the Uconnect vulnerability

          This is a dramatized retelling based on public reporting. Some moments were reconstructed from the record to convey how events unfolded; no words were ever put in the mouth of a real, named person. Everything treated as fact is drawn from government reports and first-hand journalism, cited below.

          Enjoyed this case? Follow Persist Cyber Defenders so you get the next one, and rate and review the show in your podcast app. It helps other listeners find us.

          Persist Cyber Defenders is brought to you by Persist Security: https://persistsec.com

          10 min
        • NotPetya (2017): The Most Destructive Cyberattack in History

          In June 2017 a poisoned update to M.E.Doc, a Ukrainian accounting program, unleashed a wiper that spread around the world in hours. It crippled Maersk, FedEx's TNT Express, Merck and others.

          In this episode

          • How attackers hid inside M.E.Doc's update servers for months before striking
          • How NotPetya spread with stolen passwords and why it destroyed data instead of holding it for ransom
          • How a single surviving domain controller in Ghana helped Maersk rebuild
          • Who was charged (Russia's GRU, "Sandworm") and the lessons for software supply chains

          Chapters

          00:00 — Show intro
          00:54 — Cold open: a power outage in Ghana
          01:38 — Kyiv, June 27
          03:28 — M.E.Doc and Maersk
          05:17 — How they got in: the update server
          07:22 — Spreading with stolen passwords
          09:20 — The one surviving domain controller
          11:57 — The bill
          14:12 — The human cost
          15:46 — Rebuilding Maersk
          17:58 — Who did it: GRU / Sandworm
          20:08 — Lessons
          22:33 — Closing and sources note

          Sources

          The Untold Story of NotPetya (Wired)
          US v. Andrienko et al. (GRU indictment)
          White House statement attributing NotPetya to Russia
          NCSC and partners condemn Russian cyber activity
          Maersk: Statement on cyber attack, update 3
          CISA Alert TA17-181A: Petya Ransomware
          Microsoft: Petya adds worm capabilities
          Kaspersky Securelist: ExPetr/Petya/NotPetya is a wiper

          This is a dramatized retelling based on public reporting. Some moments were reconstructed from the record to convey how events unfolded; no words were ever put in the mouth of a real, named person. Everything treated as fact is drawn from government reports and first-hand journalism, cited below.

          Enjoyed this case? Follow Persist Cyber Defenders so you get the next one, and rate and review the show in your podcast app. It helps other listeners find us.

          Persist Cyber Defenders is brought to you by Persist Security: https://persistsec.com

          24 min
        • Twitter Hack (2020): The Teenage Phone Call That Broke the Internet

          In July 2020 a group of teenagers phoned Twitter employees, talked their way into internal admin tools, and took over the accounts of Barack Obama, Joe Biden, Elon Musk and others to push a bitcoin scam.

          In this episode

          • How phone spear-phishing beat Twitter's VPN and two-factor codes
          • Why about 1,500 employees having powerful admin tools was the real weakness
          • What was taken: 130 accounts targeted, 45 used to tweet the scam
          • How it ended: the arrests, Graham Ivan Clark's sentence, and the lessons on least privilege

          Chapters

          00:00 — Show intro
          00:54 — The bitcoin scam begins
          01:29 — How they got in: phone spear-phishing
          02:27 — The bank-vault analogy
          03:19 — Who did it
          03:40 — The flood of hijacked accounts
          03:59 — The money
          04:36 — 130 accounts, 45 used
          05:18 — Response and sentencing
          05:54 — Why it worked
          06:47 — The investigations
          07:13 — Lessons
          08:33 — Closing and sources note

          Sources

          DOJ: Three individuals charged for alleged roles in Twitter hack
          Twitter: An update on our security incident (July 30, 2020)
          FBI San Francisco investigating cyber intrusion at Twitter
          Senate Select Committee on Intelligence

          This is a dramatized retelling based on public reporting. Some moments were reconstructed from the record to convey how events unfolded; no words were ever put in the mouth of a real, named person. Everything treated as fact is drawn from government reports and first-hand journalism, cited below.

          Enjoyed this case? Follow Persist Cyber Defenders so you get the next one, and rate and review the show in your podcast app. It helps other listeners find us.

          Persist Cyber Defenders is brought to you by Persist Security: https://persistsec.com

          10 min
        • WannaCry (2017): The Ransomware That Stopped the World

          In May 2017 a ransomware worm built on a leaked NSA exploit called EternalBlue tore through more than 230,000 computers in 150 countries. In Britain it forced the NHS to cancel about 19,000 appointments.

          In this episode

          • How EternalBlue turned ransomware into a self-spreading worm
          • How a 22-year-old researcher, MalwareTech (Marcus Hutchins), found the kill switch
          • What the UK National Audit Office found about the NHS's unpatched systems
          • How the US and UK attributed the attack to North Korea

          Chapters

          00:00 — Show intro
          00:54 — Cold open: the scan that won't open
          03:04 — The NHS and an unpatched weakness
          04:05 — EternalBlue
          07:05 — Ransomware plus a worm
          09:56 — MalwareTech and the kill switch
          12:13 — The accounting: 19,000 appointments
          14:16 — The human toll
          16:26 — Recovery
          18:46 — Lessons, NAO findings and attribution
          23:56 — Closing and sources note

          Sources

          UK Foreign Office attributes WannaCry to North Korea
          DOJ charges Park Jin Hyok
          UK NAO: Investigation, WannaCry cyber attack and the NHS
          NCSC: North Korea behind WannaCry
          Microsoft WannaCrypt ransomware analysis
          US-CERT Alert TA17-132A

          This is a dramatized retelling based on public reporting. Some moments were reconstructed from the record to convey how events unfolded; no words were ever put in the mouth of a real, named person. Everything treated as fact is drawn from government reports and first-hand journalism, cited below.

          Enjoyed this case? Follow Persist Cyber Defenders so you get the next one, and rate and review the show in your podcast app. It helps other listeners find us.

          Persist Cyber Defenders is brought to you by Persist Security: https://persistsec.com

          25 min
        • Stuxnet (2010): The World's First Digital Weapon

          A worm that arrived on a USB drive reached Iran's Natanz enrichment plant. It quietly wrecked centrifuges while the operators' screens kept reading "normal."

          In this episode

          • How Stuxnet spread using four previously unknown Windows flaws and stolen digital certificates
          • How it hunted for one specific Siemens controller setup and faked normal readings
          • How researchers finally pulled it apart, and what it cost Iran's program
          • Why nobody has formally claimed it, and what defenders of industrial systems learned

          Chapters

          00:00 — Show intro
          00:54 — Cold open: a USB drive and a lying screen
          03:12 — The system: centrifuges and Siemens controllers
          05:33 — How it spread: four zero-days and stolen certificates
          06:40 — The spinning-drums analogy
          08:05 — Inside the walls: through Iran's contractors
          10:22 — The machines start dying
          11:46 — The turning point: the code is pulled apart
          12:49 — The cost
          15:03 — The human toll
          17:10 — Recovery
          19:33 — A chain of trust turned against itself
          20:52 — Accountability and attribution
          22:00 — What it left behind
          24:03 — Lessons
          24:26 — Closing and sources note

          Sources

          W32.Stuxnet Dossier (Symantec)
          To Kill a Centrifuge (Ralph Langner)
          IAEA report on Iran (November 2010)
          Stuxnet and the Future of Cyber War (ETH Zurich CSS)
          The Stuxnet Enigma (Kaspersky)

          This is a dramatized retelling based on public reporting. Some moments were reconstructed from the record to convey how events unfolded; no words were ever put in the mouth of a real, named person. Everything treated as fact is drawn from government reports and first-hand journalism, cited below.

          Enjoyed this case? Follow Persist Cyber Defenders so you get the next one, and rate and review the show in your podcast app. It helps other listeners find us.

          Persist Cyber Defenders is brought to you by Persist Security: https://persistsec.com

          26 min
        • The Cuckoo's Egg (1986): A 75-Cent Error That Uncovered a KGB Spy

          In 1986 a 75-cent gap in the computer accounts at Lawrence Berkeley Laboratory set Clifford Stoll on the trail of an intruder. It led to a West German hacker ring selling stolen material to the Soviet KGB.

          In this episode

          • How a tiny billing discrepancy revealed an intruder with full control of the system
          • The GNU Emacs "movemail" flaw he used to get in
          • Why Stoll chose to watch the intruder instead of locking him out, and the "Operation Showerhead" honeypot
          • How the trail led to Hanover, the arrests, and Markus Hess's conviction

          Chapters

          00:00 — Show intro
          00:55 — Cold open: 75 cents
          03:05 — The weakness: Emacs movemail
          05:53 — The cuckoo's egg
          08:10 — Watching, not closing: printers on the lines
          10:36 — Nuclear, SDI and stealth files
          11:14 — Tracing the call to Hanover
          12:03 — Operation Showerhead: the honeypot
          13:16 — What it cost
          18:30 — Recovery and the arrest of Markus Hess
          20:20 — The conviction
          23:49 — Legacy and lessons
          26:12 — Closing and sources note

          Sources

          Stalking the Wily Hacker, Clifford Stoll, Communications of the ACM (1988)
          Stalking the Wily Hacker (archival PDF)

          This is a dramatized retelling based on public reporting. Some moments were reconstructed from the record to convey how events unfolded; no words were ever put in the mouth of a real, named person. Everything treated as fact is drawn from government reports and first-hand journalism, cited below.

          Enjoyed this case? Follow Persist Cyber Defenders so you get the next one, and rate and review the show in your podcast app. It helps other listeners find us.

          Persist Cyber Defenders is brought to you by Persist Security: https://persistsec.com

          27 min

        About Persist Cyber Defenders

        From the publisher's feed

        A hospital's screens go dark. A pipeline stops and a coastline runs dry. A bank, a phone, a password nobody thought about — and then the people who trace it back. Persist Cyber Defenders tells the…