Sriram Krishnan: Agents need fundamentally better security primitives, and we're in an awkward intermediate era
latticecut · x · 2026-10-04
Sriram Krishnan flags two warning signs of current agent usage: Claude/ChatGPT "starting to debug your browser" and agents demanding full disk or accessibility access. His argument: we lack the fundamental security primitives needed for agents to act on our behalf, and today's approaches are an awkward intermediate era rather than a real solution.
More from coding & agent
- Researchers claim $400 multi-model agent run cracked 70-year-old Pierce–Birkhoff conjecture — burny_tech · 2026-10-04
- Simon Willison: Pay-by-usage services need default hard budget caps — latticecut · 2026-10-04
- Open-source Skill lets Claude Code, Doubao and any agent call Codex's built-in image generation — vista8 · 2026-10-04
- Dev demos durable multiplayer agents running entirely on an Android phone via ngrok — itsOmSarraf_ · 2026-10-04
- Dev builds a physical control deck to juggle multiple AI coding agent sessions — Weak_Cherry_4878 · 2026-10-04
- Anthropic CEO drops 12-page PDF on using Sonnet 5.5 vs Opus 5.5 — msharmas · 2026-10-04