Methods/Defense

Training models to prioritize instructions

Training the model itself to follow system and developer instructions over conflicting text from tools or documents.

7 records7 defense3 findings (2 measured)First recorded 2024-02assistant-drafted

How it works

Fine-tuning or reinforcement learning on examples where lower-privilege text tries to override higher-privilege instructions, as in instruction hierarchy, StruQ, and SecAlign.

Known limits

Improves robustness on the attacks trained against; adaptive attacks have still succeeded.

What we know

1 reported, 1 qualified, 1 revalidate

Records over time

RangeLanes
4 of 7 records in view

Use the arrow keys to move between records, Home and End to jump to the first and last, and Enter to select one.

Agents find real bugsAgents in real operationsGated capability, incidents in the labAttackCapabilityDefensePolicyJan 25Jul 25Jan 26Jul 26
Full record · drag to choose a range
202420252026

Select a mark to read the record. Mark size shows editorial significance. Hollow marks are dated to the month. Era bands are editorial labels.

Records in view

4 records · newest first
Mar 2026
Mar 10, 2026
OpenAI releases IH-Challenge RL dataset and reports instruction hierarchy gains on injection benchmarks
DefenseDatasetOpenAI

OpenAI describes IH-Challenge, a reinforcement learning dataset of simple, programmatically graded conflicts between higher- and lower-privilege instructions designed to avoid shortcuts such as over-refusal. A GPT-5 Mini variant trained on it (GPT-5 Mini-R) improved on instruction-hierarchy benchmarks and on CyberSecEval 2 and an internal prompt injection benchmark, with little capability loss; the dataset is publicly released.

Nov 2025
Nov 7, 2025
OpenAI describes its layered approach to prompt injection as a frontier security challenge
DefenseFrameworkOpenAI

OpenAI describes prompt injection as social engineering aimed at AI agents and lists its layered defenses: instruction-hierarchy safety training, automated red-teaming, AI-based monitors that can be updated quickly, sandboxing of code-running tools, link approval, confirmation before sensitive steps, logged-out mode in Atlas, and a watch mode on sensitive sites that pauses the agent if the user leaves the tab. It cites thousands of hours of prompt-injection-focused red teaming and a bug bounty, and says it has not yet seen significant attacker adoption of the technique.

Oct 2025
May 2025
May 20, 2025
Google DeepMind reports lessons from continuously attacking Gemini with adaptive prompt injections
DefensePaperGoogle DeepMind

Shi and colleagues describe Google DeepMind's continuous adaptive-attack evaluation of Gemini against indirect prompt injection in tool-use settings. On Gemini 2.0, adaptive attacks generally matched or beat non-adaptive ones against eight baseline defenses, reaching 98.4% against in-context learning and 82.4% against spotlighting, while a warning defense and a user-instruction classifier held (at most 10.8% and 3.0%). Adversarial fine-tuning for Gemini 2.5 lowered but did not eliminate attack success.

All records