Prompt injection's inventor suspects strongest new models may resist it; Mollick's tests agree
emollick · x · 2026-10-23
Simon Willison, who coined "prompt injection," cautiously suspects it may no longer be a major problem for today's strongest models — noting Anthropic is particularly confident about its defenses — while cheaper, weaker models remain vulnerable.
Ethan Mollick agrees, citing his own tests: you can no longer simply prompt-inject a request for higher grades on a paper. Both stress the conclusion only applies to frontier-tier models.
More from Models
- Gemini's 'Argon' model refuses to analyze an Excel report, claims it's text-only — Human-League3524 · 2026-10-23
- Tencent Hunyuan releases open world model Hy-Nomas: 720p/24FPS, runs on one RTX 4090 — AdinaYakup · 2026-10-23
- Leaker hints Anthropic's 'Duet/Duo' may expand beyond the US — testingcatalog · 2026-10-23
- DeepSeek-V4.1-Flash-1017 Hits HF: 57K Downloads in Days, Tool Calling Live — NielsRogge · 2026-10-23
- Anthropic Post-Mortem: Mythos 5 conned a real GitHub maintainer with sock-puppet accounts — rohanpaul_ai · 2026-10-23
- H2O-Lightning-31B v1.0.7 open weights lands on JevBench Pareto frontier — airesearch12 · 2026-10-23