Capital, la Bolsa y la Vida · Capital Radio

Análisis de Apertura y el peligro de "Mythos"

·30 min·2 clips
The AI 'Mythos' found 27-year-old security flaws, causing an urgent US Treasury meeting with major banks.
1. Capital Radio's 'Capital, la Bolsa y la Vida,' hosted by Laura Blanco, broadcasts a live Friday morning markets analysis with a focus on Anthropic's Mythos AI model and its security implications. 2. Anchor Maite Guterres and anchor Laura Blanco are joined by Pablo Sanemeterio (CEO of Vapasec), Pablo García (Divacons Alfa Value), and Juan Esteve (Director of Investments, Markets y Zona Value). 3. The episode's central story is Anthropic's Mythos model: an AI system withheld from public release after it discovered critical security vulnerabilities and demonstrated the ability to write exploit code autonomously. 4. Anthropic's Mythos found vulnerabilities in OpenBSD — an operating system specifically designed for security — dating back 27 years, according to Sanemeterio. 5. Sanemeterio compares Mythos's exploit-writing capability to Pwn2Own, the annual hacking competition where elite researchers chain 20-30 critical vulnerabilities to gain system control. 6. The vulnerabilities are categorized into two main risks: making banking systems unavailable, and enabling ransomware or code execution within financial institutions. 7. The alarm was significant enough that US Treasury Secretary Scott Bessent convened an emergency meeting with the CEOs of Bank of America, JP Morgan, Citi, and Goldman Sachs during the same week. 8. Project Glasswinds is named as the coordinated industry response: major software providers and security firms are working together to patch all systems and flaws identified by Mythos before the model or equivalent tools are released. 9. Sanemeterio warns that while Mythos is currently contained, the speed of the emergency response suggests the problem is 'quite considerable.' 10. The asymmetric risk is highlighted: Anthropic disclosed its findings and is withholding the model — consistent with its known refusal to supply AI capabilities to the US Pentagon — but China, Russia, Iran, and Pakistan are under no such ethical constraint and will not announce equivalent capabilities. 11. Open-source AI model weights are identified as the enabling factor: any nation with sufficient compute and reasoning-model expertise could replicate Mythos-level exploit generation without public disclosure. 12. In market news, Repsol falls 2.7-4% after publishing Q1 2026 data: production of 539,000 barrels/day (down 0.2% year-over-year) and a refining margin of $10.9/barrel — strong but slightly below Q4 2025 levels. 13. Pablo García notes that $10.9/barrel refining margin is exceptional, more than doubling the prior $5.3/barrel, and attributes the stock drop partly to the broader Strait of Ormuz uncertainty. 14. French catering company Sodexo issues a profit warning, cutting underlying operating margin guidance to 3.2-3.4% from 4.39% and European sales growth to 0.5-1% from 1.5-2.5%, triggering a 17% drop in Paris. 15. Compass Group falls 13% in London, likely in sympathy with Sodexo's profit warning, given both operate in the contract catering sector. 16. TSMC (Taiwan Semiconductor) reports a 35% revenue increase for Q1 2026 with a 45% sales jump in March, lifting semiconductor stocks including ASML and Infineon. 17. Juan Esteve of Markets y Zona Value recommends Spanish banking as the best sector for investment at this moment, and flags tourism as a sector he likes but would not enter now. 18. The IBEX 35 opens at 18,118 points before drifting to slightly negative territory; Brent crude is at $97.50/barrel, gold at $4,769/oz; VIX at 20.28 signals relatively low near-term volatility. 19. Financial professionals, retail investors, and anyone tracking AI security risks or Spanish energy stocks will find the episode directly relevant. 20. Listeners seeking a deep technical briefing on Mythos, or complete coverage of any single topic, will find the live multi-topic format too fast-paced.
Listen to the show on