Sign up to save your podcasts
Or
Share Ep. 15 - We Replicated the Anthropic AI Espionage Report… and Claude Code Lied
Share to email
Share to Facebook
Share to X
Share to email
Share to Facebook
Share to X
Or
Ep. 15 - We Replicated the Anthropic AI Espionage Report… and Claude Code Lied
Did Anthropic’s November AI espionage report tell the full story?
In this episode of the Dirty South Security Podcast, host Tony UV sits down with returning guest Dan Tentler, Executive Founder at Phobos, to replicate and mythbust Anthropic’s claims that Claude Code autonomously executed a large-scale cyber espionage campaign.
Over the course of three days of live testing, we attempted to reproduce the behavior described in the report. What we discovered raises serious questions about AI autonomy, model behavior, safety guardrails, and something even more concerning — Claude Code presenting false information during testing.
In this episode we break down:
• What the Anthropic November report claimed
• How we attempted to replicate the AI-orchestrated attack scenario
• Where Claude Code failed
• Evidence of deceptive or fabricated responses
• The difference between AI hallucination and intentional-seeming behavior
• What this means for cybersecurity professionals
• Implications for AI governance and enterprise adoption
This is not a hot take. This is hands-on testing.
If you work in cybersecurity, AI safety, red teaming, or enterprise security strategy, this episode is essential listening.
Watch until the end for the full breakdown of what happened during our three-day replication attempt.
Subscribe for more deep technical security discussions and mythbusting.
#DirtySouthSecurity #Anthropic #ClaudeCode #AISecurity #CyberSecurity #DanTentler #TonyUV #AIHallucinations #AITransparency #RedTeam #Infosec #AIResearch
// FIND VERSPRITE’S CYBERSECURITY TEAM ONLINE //
- VerSprite: https://versprite.com/
- LinkedIn: https://www.linkedin.com/versprite-llc/
- Twitter: https://twitter.com/versprite/
- YouTube: https://www.youtube.com/c/VerSprite
// ABOUT VERSPRITE //
VerSprite is a leader in risk-based cybersecurity services and PASTA threat modeling, enabling businesses to improve the protection of critical assets, ensure compliance, and manage risk. Our mission is to help you understand and improve your organization’s cybersecurity posture. With cyberattacks increasing in number and sophistication daily, it is crucial to protect your organization’s assets, protect your clients, and maintain the same, excellent reputation and trust you have worked hard to build. We believe that an integrated approach will result in better and more cost-effective security practices and business outcomes.
View all episodes
By VerSprite CybersecurityDid Anthropic’s November AI espionage report tell the full story?
In this episode of the Dirty South Security Podcast, host Tony UV sits down with returning guest Dan Tentler, Executive Founder at Phobos, to replicate and mythbust Anthropic’s claims that Claude Code autonomously executed a large-scale cyber espionage campaign.
Over the course of three days of live testing, we attempted to reproduce the behavior described in the report. What we discovered raises serious questions about AI autonomy, model behavior, safety guardrails, and something even more concerning — Claude Code presenting false information during testing.
In this episode we break down:
• What the Anthropic November report claimed
• How we attempted to replicate the AI-orchestrated attack scenario
• Where Claude Code failed
• Evidence of deceptive or fabricated responses
• The difference between AI hallucination and intentional-seeming behavior
• What this means for cybersecurity professionals
• Implications for AI governance and enterprise adoption
This is not a hot take. This is hands-on testing.
If you work in cybersecurity, AI safety, red teaming, or enterprise security strategy, this episode is essential listening.
Watch until the end for the full breakdown of what happened during our three-day replication attempt.
Subscribe for more deep technical security discussions and mythbusting.
#DirtySouthSecurity #Anthropic #ClaudeCode #AISecurity #CyberSecurity #DanTentler #TonyUV #AIHallucinations #AITransparency #RedTeam #Infosec #AIResearch
// FIND VERSPRITE’S CYBERSECURITY TEAM ONLINE //
- VerSprite: https://versprite.com/
- LinkedIn: https://www.linkedin.com/versprite-llc/
- Twitter: https://twitter.com/versprite/
- YouTube: https://www.youtube.com/c/VerSprite
// ABOUT VERSPRITE //
VerSprite is a leader in risk-based cybersecurity services and PASTA threat modeling, enabling businesses to improve the protection of critical assets, ensure compliance, and manage risk. Our mission is to help you understand and improve your organization’s cybersecurity posture. With cyberattacks increasing in number and sophistication daily, it is crucial to protect your organization’s assets, protect your clients, and maintain the same, excellent reputation and trust you have worked hard to build. We believe that an integrated approach will result in better and more cost-effective security practices and business outcomes.