Playbook
Tasteful Skills
“Tasteful Skills” argues that the best agent skills are not documentation or best-practice lists.
Back to Podcast Digest
World Banks JUST got scared...
TL;DR
The IMF is now treating AI hacking as a financial stability risk, not just an IT problem — In a new warning, IMF officials said advanced models like Anthropic’s Claude Mythos and OpenAI’s “GPT 5.5 cyber attack version” could trigger systemic shocks across banks, payments, and markets.
Wes Roth’s core point is ‘skill compression’ — Tools like Mythos don’t just make elite hackers a bit better; they could give non-experts near-elite exploit capability, collapsing the time, cost, and expertise once needed to find vulnerabilities.
The real fear is scale, not one mythical super-hacker — Roth argues attackers will run 20, 100, or 1,000 agent instances in parallel against many codebases, turning cyber offense into a cost-per-exploit equation rather than a scarce human-skill bottleneck.
A cyberattack on finance can become a confidence shock before it becomes a technical catastrophe — If deposits, payroll, card payments, or mortgage systems look unreliable, markets may panic on ambiguity alone, much like the contagion dynamics people remember from 2008 and 2020.
This concern has already reached the top of global finance — Roth names Andrew Bailey, Christine Lagarde, Scott Bessent, Jerome Powell, François-Philippe Champagne, and CEOs from JPMorgan, Goldman Sachs, Bank of America, Citigroup, Morgan Stanley, and Wells Fargo as all taking the threat seriously.
The scary analogy is what happened with ChatGPT and agentic coding — Just as AI tripled ebook output on Amazon and accelerated iOS app submissions, Roth says the risk is a similar explosion in the number of capable attackers, not just modest gains for existing experts.
The Breakdown
The IMF drops a blunt warning
Wes opens with a big frame: if you have money in a bank, a retirement fund, or stocks, this matters. He explains the IMF’s role in spotting the kinds of dominoes that can turn into 2008-style meltdowns, then says the new article is explicitly about AI-driven cyber risk spilling into the financial system.
The quote that changes the conversation
He zeroes in on the IMF article, authored by senior IMF figures, and calls out the line that Claude Mythos could find and exploit vulnerabilities in every major operating system and web browser, even for non-experts. That’s the hinge of the whole video: AI risk is being described as systemic, cross-sector, and capable of concentrating failure when things go wrong.
A long sponsor detour, but with a bigger point about where AI products are heading
Roth pauses for a Victor sponsorship and uses it to make a broader point: most companies don’t need “another clever tab,” they need systems that actually take work off people’s plates. He shows Victor inside Slack and Teams doing guest research, surfacing Jensen Huang’s “We’ve achieved AGI” headline, generating interview questions, and pushing the results into Notion, all as an example of AI becoming an active co-worker rather than a chatbot.
Back to the real issue: skill compression
Returning to Mythos, Roth says critics are missing the point when they argue skilled engineers could already do this stuff. The difference, in his view, is that AI erases the old barriers of expertise, time, cost, and even language fluency, so someone doesn’t need to be a six- or seven-figure specialist to launch sophisticated attacks.
Why scale is even scarier than capability
His second big concept is scale: attackers won’t use one model instance at a time. He compares it to how people already work with coding agents in multiple tabs and cites Boris Cherny using around five tabs with sub-agents; now imagine 20, 100, or 1,000 instances hammering one codebase or many at once, until the whole thing becomes a simple cost-per-exploit problem.
From hacked banks to market panic
Roth then zooms out and says banks are not just websites with money behind them; they are the plumbing for payroll, mortgages, trading, credit, and payments. A hack against one institution can become a confidence shock, a broader hit to payments infrastructure can become a liquidity shock, and enough simultaneous disruption can become exactly the kind of market shock the IMF is warning about.
The list of people sounding the alarm is hard to ignore
He runs through names to show this is not internet hype: Canadian finance minister François-Philippe Champagne, Bank of England governor Andrew Bailey, ECB president Christine Lagarde, Treasury Secretary Scott Bessent, Fed chair Jerome Powell, and Jamie Dimon all appear in his roundup. His point is simple — when the people running global money all show up for red-alert meetings about Mythos, this has moved far beyond a niche cybersecurity debate.
The final image: AI turns mediocre hackers into a flood
Roth closes by saying the danger is not Anthropic creating one super-hacker. It’s the same pattern seen after ChatGPT and agentic coding — ebook output on Amazon tripled, app submissions surged, and now the nightmare version is thousands or tens of thousands of mediocre operators suddenly getting elite cyber capabilities and flooding the system at once.
Share
Keep Reading
Two Ways to Follow
The Weekly Echo. The inbox-shaped summary of what mattered.
New editorials announced here.
