Concern 154 · Working draft · editorial review in progress
Encoding-based jailbreaks
Encoding-based jailbreaks disguise the form of a request so that the model and its safety checks may interpret it differently.
Understanding the concern
A system may recognize meaning in altered spelling, coded text, or another representation while a filter misses it. The risk comes from a gap between what the assistant understands and what the protective layer detects. Encoding itself also has many harmless uses.
The concern is not that every encoded message is malicious. Automatically blocking unfamiliar formats can also exclude legitimate users. Evaluation should ask whether the same safety boundary survives changes in representation without assuming that one successful test covers all formats.
An example
A rule checker recognizes a prohibited message on an ordinary sign but misses the same meaning when the lettering changes.
Illustrative unless identified as a documented case.
Words in plain English
- Model
- The learned part of an AI system. It uses patterns found during training to produce an answer or prediction.
- Jailbreak
- An attempt to get an AI to act outside its intended limits.
Evidence and its limits
The study tests cipher-based conversations in specified models. Its results do not imply universal success or justify treating all encoded communication as misuse.
Yuan and colleagues · 2023 · Research study
GPT-4 Is Too Smart To Be Safe: Stealthy Chat with LLMs via Cipher ↗ (opens in a new tab)Tests cipher-based safety bypasses in specified models; the title should not be read as a universal capability claim.
Broader frameworks and background
These resources provide context. They do not establish every claim in this concern.
A question to discuss
Do safety checks follow the meaning of a request, or mainly its surface form?
Keep track of different views, what needs checking, and any next step in your own discussion notes.
Open discussion notes →Principles for the conversation
Before you finish · reflect together
Virtues and principles to consider
Virtues are qualities we practice in how we treat people and make choices. They are not scores for judging someone. Use these ideas to explore what a thoughtful response could look like.
Stewardship
Stewardship means caring for resources and systems that others depend on, including people who will use them later. It looks beyond the benefit of a quick launch.
Plan for maintenance, energy use, data protection, and eventual shutdown. Ask who will look after the tool when the original team moves on and how future users can avoid inheriting hidden costs.
Ask together: What are we responsible for caring for over time?
Prudence
Prudence means using careful judgment before and during action. It balances worthwhile opportunities with the need to avoid harm that could have been anticipated.
Start with a limited trial, set stopping points, and prefer choices that can be undone when uncertainty is high. Review what happens rather than assuming the original plan remains sound.
Ask together: What is a careful next step, and when should we stop or change course?
Accountability
Accountability means being answerable for decisions and taking responsibility for repair. It requires more than saying that the software made the choice.
Name the person or team responsible, keep useful records, and give them authority to correct problems. Make the route to a complaint clear to the people affected.
Ask together: Who will answer for the outcome and help put things right?
Editorial ratings
These are editorial judgments from the register, not measured probabilities or predictions about your situation.
About the register and scoring →More ways to explore
Search for more writing
For more specific or recent material, open a search below or ask for a web summary. AI-generated summaries can contain mistakes; open the original articles to check them.