~/talk

Scaling Smart Contract Security with AI

Talkat Pragma New Delhi 2025 · on Thursday, 2025-09-25· [security, web3]
name:Shashank
links:none on file
slides:none on file

Claims that AI-driven automation cuts static scanner noise, citing a 82% reduction in false positives after training on 100K+ audit reports and 1M+ labeled vulnerabilities. Demonstrates the Sept 2022 ShadowFi token burn exploit where rule-based scanners miss custom RBAC and public burn functions, then shows a deployed multi-agent master/slave model that embeds an “AI auditor” into CI/CD for real-time detection; current system is static analysis with plans for formal verification and AI-generated fuzzing.

AI-written summary · about

  1. 0:00Start
  2. 0:02Introduction & context: AI + blockchain security
  3. 0:27Shashank’s background in cybersecurity (Apple, Google, Facebook)
  4. 0:59Web3 security challenges: hacks & financial losses
  5. 1:28Birth of SolidityScan: AI-powered contract analysis
  6. 2:00Why automation is needed: billions lost, manual audits can’t scale
  7. 2:45CI/CD security comparison to Web2
  8. 3:09Rule-based scanning limitations → need for AI
  9. 3:27Case study: ShadowFi token hack (Sept 2022)
  10. 4:10Transaction analysis: token burn exploit explained
  11. 5:21Access control vulnerabilities as #1 cause
  12. 5:44How missing modifiers caused the exploit
  13. 6:08Why rule-based systems fail to catch custom RBAC patterns
  14. 6:59Example: false positives in static scanning
  15. 7:32Using AI to understand code context & access logic
  16. 8:05Results: 82% reduction in false positives
  17. 8:57Multi-agent AI: master/slave setup & specialized agents
  18. 9:30Dataset: 100K+ audit reports, 1M+ vulnerabilities
  19. 10:01How the AI auditor works in developer pipelines
  20. 10:25Audience Q&A: context limits, hallucinations, AI-guided algorithms
  21. 11:50Future plans: formal verification & AI-generated fuzzing tests
  22. 12:10Static analysis today, dynamic verification next
~/search
type to search 6,249 conferences, 138,987 speakers and 185,172 talks
~/help
cmdctrl k
search every conference, speaker and series
/
focus the filter search
h
home
c
conferences
s
speakers
e
series
t
light / dark — the status bar button does the same
?
this help
esc
close this, then clear every filter — or just the search, while you are typing in it

Ignored while you are typing. Pages without a filter rail have no search to focus.