korrents

← Prompt injection is unsolved, but the latest models are post-trained…

On the map

8 connected korrents · 7 moments on record from 12 Feb 2026 to 4 Sept 2026.

Same subjectSame subjectSame subjectSame subjectSame subjectSame subjectSame subjectSame subject Prompt injection isunsolved, but the latestmodels are post-trained to… PS Peter Steinberger — holds since 2026-02-12 Auto-mode does not yet convincingly fix prompt-injection risk for coding agents. Auto-mode does not yetconvincingly fix… SW Simon Willison — holds since 2026-08-08 Current prompt-injection defenses for AI agents (such as auto mode) are now reliable enough that agents can practically be assumed safe from successful injection attacks. Currentprompt-injection… ZM Zvi Mowshowitz — holds since 2026-09-04 Do not run a personal agent on a cheap or local model: weak models are gullible and easy to prompt-inject. Do not run a personalagent on a cheap or… PS Peter Steinberger — holds since 2026-02-12 AI agents were reasonable to assume a broken exploit grader would check results causally, even though it turned out not to. AI agents werereasonable to assume a… ZM Zvi Mowshowitz — holds since 2026-08-29 An open weights model from 2025 with a pentest harness could already escape a sandbox and hack most networks; the surprise says more about the sandbox than the model. An open weights modelfrom 2025 with a… TP Thomas Ptacek — holds since 2026-07-22 Frontier AI labs such as Anthropic likely have had internal security incidents similar to OpenAI's HuggingFace attack that were never publicly disclosed. Frontier AI labs suchas Anthropic likely… ZM Zvi Mowshowitz — holds since 2026-08-31 The AI models involved in the HuggingFace hack incident were severely misaligned, not merely victims of process failures. The AI models involvedin the HuggingFace hack… ZM Zvi Mowshowitz — holds since 2026-08-29 Newer model releases have got worse at producing text that is actually pleasant to read and understand. Newer model releaseshave got worse at… AR Armin Ronacher — holds since 2026-08-25
same subject