AI trust issue: models may generate plausible but incorrect content
JacquesThibs · x · 2026-08-19
JacquesThibs notes that pro-security folks believe AIs will produce groundbreaking research without bullshitting under monitoring, potentially a misunderstanding. He argues that explaining why AIs bullshit is crucial to bridging the disagreement, otherwise critics are dismissed as blind to AI capabilities.
Related event: Researchers Debate Whether Alignment Failures Are Philosophical or Prosaic(7 posts)→
More from Safety
- METR outlines three blind spots in public knowledge of AI lab internal agent risks — CFGeek · 2026-10-16
- Open-source rizin-mcp lets LLMs run local binary and malware analysis via Rizin — Pure_Emergency1424 · 2026-10-16
- METR researcher: scariest scenario is agents actively subverting their own safety systems — deliprao · 2026-10-16
- Cryptographer Matthew Green: Pull Critical Infrastructure Firmware Now — matthew_d_green · 2026-10-16
- Kernel-Level Agent Sandboxing With macOS Seatbelt, No Docker Needed — mostly_deterministic · 2026-10-16
- Stop giving agents raw API keys: prepaid balances and per-credential caps — Cold_Hall_5384 · 2026-10-16