Goldman Sachs Assesses Cybersecurity Risks of Anthropic’s Mythos AI Model
Summary
Goldman Sachs says it is working with Anthropic and cybersecurity partners to assess risks from Mythos, which Anthropic says can find and exploit sophisticated software vulnerabilities. The model is being shared with 11 organizations through Project Glasswing.
Key points
- CEO David Solomon said Goldman Sachs is closely monitoring cybersecurity risks from emerging AI models, including Anthropic’s Mythos.
- Anthropic says Mythos can identify and exploit sophisticated software vulnerabilities and may outperform human experts in some cases.
- Anthropic has restricted Mythos access and is sharing it with 11 organizations through Project Glasswing, including Goldman Sachs.
- The article says banks’ mix of modern and legacy systems may increase their exposure to AI-assisted vulnerability discovery and exploitation.
- Goldman Sachs says it is working with Anthropic, cybersecurity partners, the U.S. government and model publishers to strengthen cyber and infrastructure resilience.
Article Details
- Event Type
- Financial-sector evaluation of cybersecurity risks from Anthropic’s advanced AI model Mythos, including controlled access through Project Glasswing.
- Impact
- Mythos is reported to autonomously discover and exploit sophisticated software vulnerabilities, potentially increasing risks to financial systems. Goldman Sachs is monitoring the risks and working with Anthropic and cybersecurity partners to strengthen defenses. The article reports potential risks, not a cybersecurity incident or realized impact.