Official Website link
Date: October 9, 2026
Location: Grand Hyatt, Bolgatty, Kochi, Kerala
Event: c0c0n 2026 Conference
Format: Conference Talk
Speaker: Anant Shrivastava | Founder, Cyfinoid Research
Overview
The CFP for this talk was submitted in May 2026 with a Mythos readiness focus. By October the year had supplied its own material: Shai Hulud and Team PCP categorically proved that the development pipeline starts at the desktop rather than the CI system, the Mythos launch and its guardrailed sibling Fable reshaped the model landscape, Project Discovery established that an open weight model can be backdoored for as little as 50 dollars, and OpenAI’s disclosures about unauthorized agent collaboration turned AI-initiated attacks from a theory into incident reports.
The talk works through three different ways the same system gets attacked: human attackers using AI, AI acting on its own, and AI adopted by organizations. For people who are mere users of AI, both practitioners and executives, the practical core is a readiness framework of ten checklist areas (A1 to A10) covering asset visibility, identity and privilege, patch speed, detection coverage, response authority, recovery confidence, developer and CI/CD governance, backlog reduction, safe internal AI use, and defensive AI automation. Each area comes with a headline question and verifiable evidence instead of vague maturity claims.
The talk closes with defensive automation principles: transition probabilistic AI output into deterministic controls by exploring, assisting, automating, encoding, and verifying.
Key Topics
- 2026 in review: Shai Hulud, Mythos and Fable, model backdooring, unauthorized agent collaboration
- Three attack paths through one system: humans using AI, autonomous AI, organizational AI
- What AI changes offensively: faster discovery, exploitation, and weaponization, with no breaks
- Practitioner readiness checklists A1-A10 with headline questions and verifiable evidence
- Defensive automation: moving from probabilistic AI output to deterministic controls
