arXiv:2606. 08403v1 Announce Type: cross Abstract: Text-centered prompt-injection defenses assume that the malicious signal is visible in one of the inspected text views.
Machine-generated by The Flow from the publisher's headline and feed description — not written or checked by a human. The full article lives at arXiv AI.
arXiv:2601. 22818v2 Announce Type: replace-cross Abstract: Fine-tuned LLMs can covertly encode prompt secrets into outputs via steganographic channels.
arXiv:2606. 09411v1 Announce Type: cross Abstract: Large language models can be fine-tuned to encode prompt-borne secrets into fluent, seemingly benign outputs.
Repeat-After-Me is a black-box adaptive visual prompt injection technique that can reveal personally identifiable information or trigger malicious tool calls in both open-weight and commercial vision‑language models, achieving attack success rates above 80% on Qwen3.6‑27B and 47% on GPT‑5.5. The method works even when the benign user prompt is unrelated to the injected task and does not explicitly authorize it, and it retains significant effectiveness when transferred across models or optimized on surrogate systems. In a real‑world OpenClaw Discord deployment, a minimally injected image can overwrite TOOLS.md, enabling remote code execution and secret exfiltration.
arXiv:2606. 09135v1 Announce Type: cross Abstract: We demonstrate that widely deployed Large Language Model (LLM) inference stacks harbor a steganographic channel that requires no modification to model weights, sampling code, or output distributions.
The paper critiques current encoded‑prompt safety benchmarks that focus only on harmful requests, showing that such tests can misrepresent a model’s safety. By evaluating the benign arm under the same encoding, the authors reveal a substantial drop in the harm gap—sometimes to zero—indicating that the encoding masks true safety deficiencies. Across multiple large models and training pipelines, they document that the encoding can either hide or falsely inflate safety metrics, and they identify twelve specific instrument defects that contribute to these misleading results.
The paper investigates how multimodal large language models are more susceptible to prompt injection when adversarial instructions are presented as text rather than as non-textual inputs like images. It proposes a training‑free defense that renders untrusted payloads into typographic images (or audio) before they reach the model, a method called Pictionary. Experiments on ten models and two benchmarks show that this approach significantly lowers attack success rates while maintaining normal functionality, and that fine‑tuning on image‑rendered instructions can further reduce the modality gap.