Won't Fix

Juliet Shen is cofounder and Head of Product at ROOST (Robust Open Online Safety Tools), a nonprofit building open-source trust-and-safety infrastructure for platforms of every size. She has done anti-abuse product work at Google and Grindr, and was the first trust-and-safety product manager at Snap, where she helped launch early cross-platform efforts to combat child exploitation. ROOST's website is https://roost.tools.

Across her career, Juliet kept running into the same maddening pattern: every trust-and-safety team, at every platform, quietly rebuilding the same rules engines, review queues, and reporting pipelines from scratch, behind closed doors, and at enormous cost. ROOST is a bet that online safety should be shared, open infrastructure rather than proprietary secret sauce, available free to any platform or site that needs it. We talk about that, and a lot more.

Key Highlights:

• Every tech company shouldn't have to build their trust and safety tools from scratch behind closed doors. It’s an expensive waste of time when open-source infrastructure could solve the exact same foundational problems for everyone.
• PMs and engineers need to step up and lead in the trust and safety space. They are the ones who can actually bridge the gap and get policy, operations, engineering, and legal teams talking to each other.
• AI is great for knocking out the easy, baseline moderation tasks. But when a situation is highly nuanced or something the AI hasn't seen in its training data, you still absolutely need human judgment.
• As social media breaks apart into decentralized networks, a one-size-fits-all safety system won't work anymore. We need modular tools that let platforms look at who the user is, how they're behaving, and what they're posting as separate pieces of the puzzle.
• Good moderation is often less about analyzing the post itself and more about knowing exactly who is behind the account or the app. Right now, our lack of solid identity verification is a massive blind spot for digital safety.

Chapter Timestamps:

00:00 Introduction

1:34 Career Journey and the Problem of Redundant Tool Building

5:30 The Role of Product Managers in Trust and Safety Teams

7:34 Impact of LLMs on Trust and Safety Operations

11:27 Focus Areas and Child Safety Priority

12:55 The ABC Framework and Actor Trust Challenges

16:54 Community Building and TrustCon Participation

19:13 Signal Sharing vs Tool Sharing Philosophy

22:43 Open Source Approach and Scaling Challenges

23:59 Future Roadmap and Research Partnerships

28:52 Reviewer Well-being and Mental Health Considerations

32:00 Centralized vs Decentralized Moderation Models

37:15 Government Role and Open Source Support

39:52 Success Metrics and Measurement Challenges

42:50 Standards, Testing, and Future Directions

Resources & Links:

Rob Leathern (https://www.linkedin.com/in/leathern/)

Juliet Shen (https://www.linkedin.com/in/julietshen/)

ROOST (https://roost.tools)

Dave Kleidermacher is a vice president of engineering at Google, leading engineering for Android security and privacy. His scope encompasses Android and the Made-by-Google world — Pixel, Nest, Fitbit, and the Play Store.

We talked about Android's answer to scams: smarter defenses that use AI as a shield (on-device detection that catches scams as they unfold), and a deeper structural pivot to "Actor Trust" — establishing provable, cryptographic confidence in who or what a source is rather than forever trying to detect bad things.

Dave has been steeped in these topics for a long time so we get into a bunch of great territory, and I think you’ll really enjoy the conversation.

Key Highlights:

• Consumer platforms must pivot from traditional vulnerability exploitation defenses to fighting scams and fraud, which make up 99% of actual practical threats facing users today.
• The future of mobile authentication lies in reversing security asymmetry through "actor trust" cryptographically verifying the source device rather than relying on human intuition.
• Big Tech players like Apple and Google need to publish a transparent, accountability driven joint priority roadmap to accelerate cross-platform security for critical defenses like caller verification.
• Mobile network operators remain a critical structural weak point in consumer safety due to privacy-invasive habits like silent third party app installations and outdated location-tracking protocols.

Chapter Timestamps:

00:00 Introduction and Background

3:01 The Shift from Vulnerability Threats to Scam Prevention ‎

6:01 Real-time Voice Spoofing Capabilities and Demonstrations ‎

8:19 Platform Defense Strategies and the Whack-a-Mole Problem ‎

11:00 Actor Trust and Cryptographic Verification Approach ‎

15:37 Google's Security Key Success and Developer Ecosystem Verification ‎

17:35 RCS Standards and Industry Collaboration Challenges ‎

27:19 Business Caller Verification and Stir Shaken Limitations ‎

31:59 Privacy-Security Balance and Binary Transparency ‎

41:30 Consumer Role and Stakeholder Responsibilities ‎

43:27 Future AI Landscape and Industry Recommendations ‎

49:47 Advertising Technology and Platform Accountability ‎

Resources & Links:

Rob Leathern (https://www.linkedin.com/in/leathern/)

Dave Kleidermacher (https://www.linkedin.com/in/davekleidermacher/)

My guest today is Jeremy Galen, founder of Charlemagne Labs. Jeremy spent twelve years at Meta working in privacy, safety, and security — most recently five years as a product manager in trust and safety, focused on machine-learning content enforcement, account access, impersonation, and plagiarism.

He left to start Charlemagne Labs, a New York startup building what he calls a "digital bodyguard" — an on-device AI assistant, Agent Charley, that steps in before a worker clicks a dangerous link or pastes sensitive data into a chatbot.

The company's research recently landed in Meta's safety report for its frontier model, Muse Spark, where Charlemagne's benchmark measured how capable leading AI models are at multi-turn social engineering. His core argument is that the old "think before you click" model of security is broken, and that risky digital behavior should be treated less like a moral failure and more like a public-health and system-design problem.

Learn more about Jeremy and the company at https://charlemagnelabs.ai/

Listeners who sign up for the Pro plan can get 6 months for free if they use the promo code ROB2026.

Key Highlights:

• Selling consumer security software is a non-viable market because consumers buy what they want, while businesses buy what they need.
• The open internet operates as an active battlefield where users face direct threat vectors from sophisticated foreign adversaries.
• Falling for social engineering scams is entirely situational, rather than a reflection of an individual's intelligence.
• Real-time, automated AI interventions are far more effective at enforcing digital hygiene than relying on static digital literacy training.
• Over 90% of modern cybersecurity incidents originate from human risk vectors where an individual is directly targeted or manipulated.

Chapter Timestamps:

00:00 Introduction and Guest Background

1:02 Career Transition and Startup Journey

2:33 Consumer vs. Business Security Market Analysis

3:56 Personal Motivation and Scam Prevalence

5:09 Social Engineering Sophistication and Victim Blaming

8:01 Big Tech vs. Startup Challenges

13:59 Fundraising Reality and Survivor Bias

18:05 Digital Hygiene and AI-Powered Protection

22:06 Privacy-First Architecture and Local Models

28:18 Democratizing Security and Luxury Concerns

31:59 Meta Collaboration and Industry Standards

35:16 Founder Advice and Problem Selection

38:08 Company Information and Target Market

Resources & Links:

Rob Leathern (https://www.linkedin.com/in/leathern/)

Tate Jarrow is the Founder and CEO of Rebound (https://trustrebound.com), a consumer anti-scam company. Before founding Rebound, Tate was an Army infantry officer and Airborne Ranger, and then a Special Agent at the U.S. Secret Service.

At Google, he helped start a company called Beacon through the Area 120 incubator, which was then acquired into Google One.

Key Highlights:

• What Rebound is building: "Antivirus but for scams" — software that sits on a user's device across macOS, Windows, iOS, and Android, sees what the user sees, and alerts when it detects an inbound scam. Currently in alpha, heading into paid beta within the month, with general availability targeted for summer.
• Why now: Normal people have zero real defense against scams. Law enforcement don't have resources for individual cases, and platforms are hard to reach for recovery. Existing consumer cybersecurity is rooted in 20-year-old problems (antivirus, credit monitoring) and isn't built for AI-powered, personalized, scaled attacks.
• “You can't arrest your way out of cybercrime”: Cyber criminals run transnational organizations as businesses with P&Ls, so the real lever is changing the economics.
• Google: Tate started in legal/investigations chasing cybercrime actors on Google platforms, got frustrated by the gap between business incentive and what could actually be done. Two of his Area 120 teammates are now on the Rebound team.
• Scam overconfidence: Tate shares that a GASA study found the #1 predictor of being scammed is confidence that you can spot one — overconfidence is the actual risk factor. Every demographic gets hit.
• Regulation and data: US regulation is 20 years behind. The real risk now is social engineering powered by leaked addresses, phones, emails, and contacts. He wants companies held accountable for the social engineering risk they create, not just PII in the narrow legacy sense.
• "Caring guardians": People in tech are the de facto security help desk for their parents, friends, and families. Rebound is building features so a tech-savvy family member can have visibility into risk across the people they care about — plus in-app trust verification (one-click identity check) for the "is this actually my friend messaging me?" problem.

Chapter Timestamps:

00:00 Introduction and Background

1:26 Rebound's Mission and Product Overview

3:39 Technical Implementation and Current Status

4:45 Motivation Behind Consumer Protection Focus

7:45 Google Journey and Area 120 Experience

14:59 Law Enforcement Perspective on Cybercrime

18:30 Evolution of Cybercriminal Organizations

21:07 Current State of Consumer Protection

30:04 Regulatory Environment and Government Role

37:25 Community Protection and Cross-Platform Challenges

43:00 Product Vision and Future Plans

Resources & Links:

Rebound (https://trustrebound.com)

Tate Jarrow (https://www.linkedin.com/in/tatejarrow/)

Rob Leathern (https://www.linkedin.com/in/leathern/)

Jonathan spent thirty years inside the machine — product leadership at DoubleClick, executive roles at Google, and a founding role at Harvard's Applied Social Media Lab. A year ago, Jonathan started writing Platformocracy, a newsletter with a simple, uncomfortable thesis: tech companies didn't set out to govern us, but they do now. Billions of people are subject to rules they didn't vote for, enforced by systems they can't see, with no meaningful right of appeal — built, in many cases, by people who genuinely wanted to do the right thing.

We talk about how that happened, what it looks like from the inside, and the question that may define the next five years: what happens when AI floods every platform with infinite synthetic content, and the only thing standing between us and the noise is an algorithm the noise was engineered to exploit?

Key Episode Takeaways:

• The Governance Illusion: Tech platforms have evolved into unelected global governments that impose top-down rules on billions of users who have zero democratic input or meaningful right of appeal.
• The Category Mistake: Treating platforms strictly as private businesses that can refuse service ignores the reality that they host deeply rooted human communities where "exiting" the platform means abandoning essential real-world relationships.
• Decomposing Social Media: Effective regulation requires breaking "social media" down into three distinct product categories—media consumption, community networking, and creator relationships—because a blanket approach fails to address the unique harms of each.
• Shifting the Regulatory Burden: Instead of forcing mass identity verification, regulators should require platforms to accept enhanced safety obligations and standardized parental controls if they choose to profit from serving children.
• Inverted Safety Baselines: Unlike heavily regulated sectors like automotive or food hospitality, tech platforms operate on a model where they maximize user safety only up to the point that it threatens their profit margins.

Episode Highlights:

00:00 Introduction

1:43 The Challenge of Corporate vs. Community Framing

2:54 The Evolution from Community Management to Corporate Governance

20:48 Age Verification Concerns and Technical Challenges

24:47 Historical Context and Generational Perspectives

27:46 AI, Anonymous Accounts, and Platform Integrity

37:19 AI's Potential for Improved Parental Controls

42:40 Regulatory Approaches: Enhanced Obligations for Serving Children

45:18 Profit vs. Safety Standards in Tech Industry

50:43 Procedural vs. Substantive Law in Platform Governance

Links:

Read Jonathan's newsletter: https://www.platformocracy.com

Jonathan Bellack

Rob Leathern

Craig Silverman is an award-winning journalist who has spent more than 15 years researching and reporting on the manipulation of our information environment. He is currently the co-founder of Indicator, a media outlet dedicated to exposing digital deception and teaching digital investigative and OSINT (open-source intelligence) techniques.

Prior to launching Indicator, Craig was a national reporter at ProPublica, where he focused on investigating digital platforms and online manipulation. Before that, he served as the media editor for BuzzFeed News, where he pioneered innovative approaches to exposing digital disinformation and media manipulation.

Key Episode Takeaways:

• The Industrialization of Deception: Digital manipulation has shifted from lone actors into a massive, industry backed by venture capital and brutal supply chains, including Southeast Asian "scam compounds" that merge human trafficking with high-tech fraud.
• The "Manufactured Organic" Loophole: Brands are now using "clipping" and industrial-scale UGC campaigns to generate billions of views through paid creator networks that mimic authentic posts.
• An Incentive to Cheat: The current digital economy creates a "race to the bottom" where deceptive or violative content often sees higher engagement and lower costs than honest ads.
• Ad Revenue Cannibalization: By failing to police undisclosed marketing, social platforms are letting a shadow ad economy thrive that actively drains budgets away from their own official, trackable ad businesses.
• Deterrence Through Public Examples: Instead of trying to automate everything, platforms could flip the script by making high-profile, public examples of agencies that openly brag about their deceptive tactics on social media.

Episode Highlights:

00:00 Introduction and Background of Craig Silverman

01:21 Early Collaboration and Scam Evolution

04:27 Indicator Media's Mission and Approach

08:29 Undisclosed Marketing and UGC Campaigns

13:21 Scale and Enforcement Challenges

20:51 Platform Cannibalization and Business Impact

28:29 AI Labeling Audit Results

34:15 Community-Based Detection and User Skills

39:17 Affiliate Marketing Case Study

46:48 Systemic Incentive Problems

49:13 Conclusion and Resources

Links:

Craig Silverman

Indicator

Rob Leathern

Assaf Kipnis spent years hunting financially motivated bad actors on Meta's e-crime team and in Google's Ads Trust & Safety org.

He now runs KTLYST Labs, where he's building the threat intelligence tooling he always wished existed inside big platforms. We get into the practical realities of scam fighting — what's actually changed in the AI era, what hasn't, and why so much of the industry's effort gets aimed at the wrong targets.

About the guest:

Assaf Kipnis is the founder of KTLYST Labs. Previously: Meta e-crime, Google Ads Trust & Safety, ElevenLabs, LinkedIn threat intel.

What We Cover:

• Why AI isn't reinventing scams — it's just adding a more convincing final layer to playbooks that have existed for years.
• The asymmetry problem: bad actors run conferences, sell each other tools, and share playbooks on Telegram, while defenders can't share findings across teams at the same company.
• A case study in what actually works — how changing product, policy, and operations together pushed a misinformation-for-profit ring off the platform in a week.
• Why "accounts taken down" is a near-useless metric, and the "learned futility" it creates inside big trust & safety orgs.
• The Swiss cheese model of abuse prevention, and why chasing a single silver-bullet solution keeps companies chasing their tail.
• Where regulation has teeth (banking) and where it's mostly performative (social media), plus the cross-platform gap no one is addressing.
• How AI is changing investigative work — compressing a week of open-source research into two hours — and why that makes entry-level talent pipelines a real concern.

Episode Highlights:

00:00 Intro

01:06 Professional Background and Career Journey ‎

03:47 AI's Role in Scaling Rather Than Changing Scams ‎

07:05 Adversary Collaboration vs. Defender Silos ‎

09:02 The Frame Rate Discovery Example ‎

10:26 KTLYST Labs and Operationalizing Threat Intelligence ‎

12:40 AI's Impact on Investigation Work ‎

15:15 Career Entry Points and AI's Impact on Junior Roles ‎

20:38 The NextTag Affiliate Program Attack ‎

23:00 The Misinformation Campaign Investigation ‎

27:52 The Limitations of Location-Based Solutions ‎

30:30 The Futility of Single-Solution Thinking ‎

33:47 The Reality of Platform Defense Goals ‎

34:50 Government Regulation and Enforcement Challenges ‎

40:31 The Problem with Takedown Metrics ‎

Links:

Assaf Kipnis

KTLYST Labs

Rob Leathern

Investigative journalist Tom Wright (Project Brazen) joins Rob Leathern to discuss the staggering rise of Benjamin Mauerberger, a South African money launderer who utilized crypto exchanges and high-level political "state capture"; to fund a billionaire lifestyle of super yachts and private jets while evading an international dragnet.

Tom shares more about the dark underbelly of a $200 billion global scam industry where industrial-scale "pig-butchering" complexes in Southeast Asia target citizens around the world.

Key Episode Takeaways:

• The "state capture" playbook enables global fugitives: Large-scale money launderers use their wealth to gain political protection, setting national digital policies and even attending cabinet meetings to integrate criminal proceeds into traditional banking systems.
• Cryptocurrency serves as a high-speed financial superhighway: Modern fraud has moved beyond traditional banking into crypto "piping," allowing scammers to move value across borders with frictionless speed and scale.
• A "double victimization" cycle defines the scam industry: The global fraud network relies on a brutal labor model where workers are often human trafficking victims lured by legitimate job offers only to be imprisoned and tortured within scam compounds.
• Economic impact now rivals Fortune 500 revenues: Estimates suggest the US economy loses approximately $200 billion annually to these scams—a figure that exceeds the annual revenues of automotive giants like GM or Ford.
• Jurisdictional arbitrage creates a "cat and mouse" regulatory game: Criminal entities constantly shift operations to less regulated territories, such as moving from the Seychelles to the Turks and Caicos, to evade tightening anti-money laundering oversight.
• Reputational "whitewashing": Questionable financial entities attempt to gain mainstream legitimacy by sponsoring world-class athletes or prestigious events to obscure their underlying involvement in global money laundering networks.

Episode Highlights:

00:00 Intro

01:45 The Genesis of Billion Dollar Whale

04:47 Evolution from Traditional Fraud to Crypto-Enabled Scams

08:13 Mauerberger's Rise and Political Connections

10:35 Mauerberger's Flight and Current Status

15:48 The Crypto Money Laundering Operation

24:40 The Human Cost and Complexity of Scam Operations

27:14 Challenges in Reporting and Government Response

32:34 The Broader Implications and Future Outlook

Links:

• Project Brazen
• Billion Dollar Whale
• Tom Wright
• Rob Leathern

In this first episode of Won't Fix, Rob Leathern talks to Jason Zoltak.

Jason is the founder and CEO of tofu, which is using AI and machine learning to fight fraud and deception in hiring and recruiting.

About Won't Fix: In software engineering, “won’t fix” describes a bug by acknowledging the issue but intentionally leaving it unsolved because addressing it is too costly, risky, or not worth the trade-offs.

Hear from the practitioners fighting phishing, deepfakes and bots, and learn about the broken systems and misaligned incentives that keep us all vulnerable.

Key Episode Takeaways:

• The Identity Fraud Pivot: tofu shifted from an AI resume screening tool to a fraud detection platform after discovering that remote hiring has enabled a massive surge in sophisticated identity misrepresentation.
• Near-Universal North Korean Infiltration: Virtually every company hiring for remote technical roles is now a target for North Korean IT workers, with some applicant pipelines reaching 80% fraud rates.
• The Fragmentation Vulnerability: The lack of a "digital passport" and the break in verification when moving a candidate from LinkedIn to an internal ATS creates a massive security gap for fraudsters to exploit.
• Shift in Security Ownership: Candidate fraud is transitioning from a Talent Acquisition burden to a CISO priority as companies realize recruiters lack the budget and expertise to fight organized cybercrime.
• Economic Scalability of Fraud: Fraudsters aren't looking for long-term tenure; they use deepfakes and proxies to "job stack," collecting multiple salaries simultaneously for a few months before being caught.
• The "Confirmation Bias" Trap: Once a candidate reaches the final interview stages, hiring managers and recruiters are psychologically prone to ignore red flags, making them vulnerable to sophisticated identity theft.

2:29 Jason's Background and tofu's Evolution

4:09 Discovering Candidate Fraud Through Direct Investigation

5:04 Market Response and Business Pivot Decision

6:35 Personal Motivation and AI Identity Challenges

8:17 Spectrum of Fraud vs. Embellishment in Hiring

10:25 Prevalence of North Korean IT Worker Infiltration

11:30 Evolution of Fraud Techniques and Identity Theft

13:18 Root Causes: Platform Disconnection and Identity Verification

15:26 Security vs. Talent Acquisition Budget and Responsibility Issues

17:36 LinkedIn Verification Challenges and Behavioral Incentives

19:20 Impact of Thin Digital Footprints on Legitimate Candidates

21:35 False Positive Management and Digital Footprint Requirements

24:16 Interview Process Fraud: Deepfakes and Proxy Detection

26:01 Sophisticated Deepfake Case Study and Technical Evidence

28:17 Economic Incentives and Scaling Strategies for Fraudsters

29:26 Corporate Espionage and Strategic Target Selection

32:15 Recruiter Incentive Conflicts and Trust Erosion

36:13 Critical Case Study: Final Round Interview Fraud Detection

37:28 Government Regulation vs. Private Sector Solutions

39:39 Upcoming Product Launches: ATS Reconnaissance and Continuous Monitoring