Thoughtworks Technology Podcast Anthropic Mythos: Hype, reality and the actual security implications
49 snips
Apr 30, 2026 
Chris Kramer, AI and machine learning leader at ThoughtWorks, gives a concise mini bio and frames Mythos as a stepwise advance rather than a doomsday AI. He and Nate unpack Project Glasswing, how Mythos surfaced long‑standing bugs, agentic security scans, risks of agentic pipelines, and the organizational and architectural shifts shaping where breakthroughs may come next.
AI Snips
Chapters
Books
Transcript
Episode notes
Mythos Is A Stepwise Capability Leap
- Mythos is not a doomsday AI but a stepwise advancement that unlocks hidden model capabilities and scales reasoning without massive new hardware.
- Chris Kramer explains Anthropic unlocked latent reasoning by adding loops into model reasoning, increasing capability without huge chip upgrades.
Mythos Accelerates Vulnerability Discovery
- Mythos finds real security bugs at speed but many would likely be discoverable by older models given time and the right agentic setup.
- Nate Schuda and Chris Kramer note Mozilla fixed hundreds of bugs from Glasswing, highlighting enterprise code's long-standing brittle spots.
Integrate Agentic Security Into CI/CD
- Do embed AI security scans and agentic checks into CI/CD pipelines as part of a developer platform to catch vulnerabilities early.
- Chris Kramer recommends agent-first codebases with automated scans, but warns of document poisoning on long-running agent tasks.
