Keynote: The Defender Advantage Has a Single Point of Failure
summary
Claim: within 2–5 years CI/CD pipelines run Mythos-class models so products stop shipping exploitable security bugs. That permanent defender advantage has a single point of failure: the model weights — if they leak, advantage inverts overnight. Open-weights parity arrives in about 7–10 months by the speaker’s estimate, so defenders must adopt zero trust, AI-driven detection and automated remediation now. Confidential computing and TCBs enable secure, global model deployments and controlled defender access.
AI-written summary · about
chapters (7)
- 0:00Securing frontier AI
- 1:14Emerging threat landscape
- 4:06The impact of Mythos
- 6:41The permanent defender advantage
- 10:17Timeline for defense
- 11:47A defender's playbook
- 15:06Confidential computing's role
related talks
16mKeynote: Defending AI on Kubernetes: Red Team Playbook
18mCommit Secure AI: Embedding Security in Code, Pipelines, and Production
13mKeynote: Software Security: _You_ are the Calvary
26mObjection! AI Security Mistakes on Trial With Kubeflow and Confidential Computing
43mKeynote – Every Threat Model is Wrong
15mKeynote
32mSoftware security as a force of nature
36mKeynote: The Future of Software