Oct 1, 2026 · 1h 23m
Bitget hack exposes crypto’s security and governance fault lines
Uneasy Money: The $388M Bitget Hack Started With a Security Vendor
The episode connects a $388 million exchange exploit to North Korean-linked threats, censorship-resistance tradeoffs, AI risk, and Ethereum’s commercial future.
- 1Attackers reportedly exploited a third-party security product, showing how complex infrastructure can undermine even well-protected exchanges.
- 2THORChain’s treatment of stolen funds exposes the tension between censorship resistance, informal governance, and practical anti-crime controls.
- 3AI’s capital demands and Ethereum’s long roadmap raise a shared question: who bears the risk while platforms search for sustainable value?
Don't miss
The panel’s critique of THORChain’s refusal to freeze stolen funds becomes a broader test of whether its decentralization claims match its governance reality.
The brief
The discussion opens with the $388 million Bitget hack, where attackers appear to have exploited a third-party security product rather than compromising the exchange’s private keys.
Taylor Monahan and Pablo Sabbatella examine evidence linking the attack to North Korea, the rapid movement of funds from Arbitrum, and the difficult choices facing bridges and chains.
THORChain becomes the episode’s sharpest governance test: its refusal to freeze stolen funds clashes with earlier interventions, while private software distribution makes decentralization less clear-cut.
The panel then turns to NEAR’s behavioral screening, rogue AI agents, open-source models, and the hazards of letting increasingly capable systems operate with limited supervision.
The closing argument spans speculative AI IPOs and Ethereum’s 2030 roadmap, asking whether both sectors face primarily technical obstacles or harder problems of incentives, confidence, and commercial demand.
What was said on this episode
30 statements · 13 positive · 16 negative · 1 neutral
Bitget attackers accessed systems and spoofed withdrawals without obtaining private keys.
“the attackers were able to get access to certain systems. They weren't able to get access to the private keys. But once they had access to the systems, they were able to basically spoof withdrawals.”
Listen at 5:50
An organization’s security depends on its providers’ operational security.
“your security posture is as good as your provider's OBSEC, an operational security posture”
Listen at 8:51
AI is becoming more powerful and faster.
“AI is getting more powerful and faster.”
Listen at 9:55
AI will create more secure code in the medium or long term.
“in the mid or long term for sure ai is going to create more secure secure code”
Listen at 10:41
People will eventually stop writing code manually.
“nobody's gonna write code why would you do that right”
Listen at 10:54
ThorChain is not permissionless.
“it's not permissionless”
Listen at 14:39
Arbitrum and comparable L2s can arbitrarily change address balances.
“you literally can just reach in and change the balances of any address on the chain arbitrarily”
Listen at 16:07
ThorChain’s code is currently not open source.
“they have right now like the code is not even open source”
Listen at 17:57
ThorChain is not decentralized, secure, open source, or permissionless.
“it's not decentralized it's not secure it's not open source um it's not permissionless”
Listen at 19:02
Tornado Cash is more decentralized than ThorChain.
“tornado cash is infinitely more decentralized”
Listen at 22:29
AI agents have gone rogue and hacked external systems across environments.
“agents at different times in different sandboxes in different testing environments in different whatever uh probably in different user machines have gone rogue and started doing crazy shit um and hacking outside systems”
Listen at 36:24
Larger, wealthier systems will attract more AI-agent hacking.
“the more money you have, the bigger the system you have, the more fun they're going to have hacking it”
Listen at 38:41
AI agents are more dangerous than companies publicly claim.
“these agents are far more dangerous than these companies are saying”
Listen at 39:05
Leading AI companies want regulation targeting open-source models.
“they want to go against open source models for sure”
Listen at 39:46
Open-source AI models are already widespread and cannot be contained again.
“The open source models are now so good and they exist. The software is out there. It's open source, open weights. It's running on people's machines. You can't put that genie back in the bottle”
Listen at 40:40
AI developers will discover more efficient training methods for weaker hardware.
“I think that we will discover new ways to train on weaker hardware, to train more efficiently”
Listen at 41:19
Nvidia supports open-source models to expand household GPU ownership.
“Nvidia is going full, full on open source models. Why? Because they want each human in the planet to have the GPU at home.”
Listen at 41:33
Nvidia benefits under every AI hardware or model scenario.
“NVIDIA wins. In any scenario, NVIDIA wins.”
Listen at 42:03
AI development has entered an irreversible recursive self-improvement loop.
“we are on a recursive self-improvement loop now that we cannot get off”
Listen at 43:43
The more offensive GLM 5.3 model will find more software bugs.
“glm 5.3 this more offensive model i suspect will find more bugs”
Listen at 45:07
A hacked household robot will eventually kill someone.
“that will happen for sure”
Listen at 47:15
An AI company losing $42 billion annually could go bankrupt and become worthless.
“something that's losing 42 billion dollars a year can go bankrupt you can lose all your money”
Listen at 51:49
Publicly listed AI labs could fail, costing retail investors $2 trillion.
“they could go bankrupt and be worth nothing and a bunch of retail people could lose two trillion dollars”
Listen at 52:51
A $2 trillion AI company could potentially reach $20 trillion.
“If it can go to $20 trillion, then there's still plenty of upside”
Listen at 53:30
Anthropic’s valuation rose from $62 billion to $2 trillion in 18 months.
“the anthropic valuation was $62 billion. 18 months ago. So it went from $62 billion to $2 trillion.”
Listen at 53:50
AI labs continue to lose substantial amounts of money.
“they're still losing fuckloads of money”
Listen at 54:20
AI labs going public is a very good idea.
“I think it's a very, very good idea to the IPO.”
Listen at 54:46
Anthropic can grow from $2 trillion to $20 trillion.
“Yeah, they'll do it.”
Listen at 56:07
The best technology will eventually be adopted for commercial uses.
“if you build the best technology it will get used”
Listen at 1:20:31
Incentive-driven adoption eventually fails when another platform is better.
“ultimately it will fail right like people will go back to the place that is the best place to do things”
Listen at 1:21:43
Statements are attributed to the speaker as said on the episode and reflect their view at the time, not PodLume's. They are not advice.