Advanced
AI Alignment & Safety
CommunicationSafety reportsRisk
Communicating AI Safety Findings — Language
5 exercises — Practice the language used in AI safety reports, capability disclosures, and risk tier communication.
0 / 25 completed
1 / 25
A safety team says: "This feature imposes an alignment tax." What do they mean?
The 'alignment tax' is the practical cost of making a model safer — e.g., refusals that reduce helpfulness, extra inference latency from safety classifiers, or reduced benchmark scores. It is a real trade-off AI teams must weigh.
2 / 25
A researcher discovers a jailbreak in a deployed model. Following responsible disclosure practice, they should:
Responsible disclosure in AI mirrors cybersecurity practice — the discoverer notifies the developer privately first, giving them time to patch or mitigate before the vulnerability becomes public knowledge.
3 / 25
Which sentence correctly uses dual-use in an AI context?
Dual-use in AI safety means a capability can serve beneficial purposes (e.g., biosecurity research) but could also be misused (e.g., bioweapon design). Dual-use capabilities require extra scrutiny before deployment.
4 / 25
When a capability is classified as a high risk tier, this means:
Risk tiers classify AI capabilities by potential harm (e.g., CBRN risks, cyberoffense, CSAM) and determine what mitigations must be in place before a model with that capability is deployed or made accessible.
5 / 25
Fill in the blank: "We identified a jailbreak vector and are implementing a ___ in the next model version."
A 'mitigation' is the standard term for a countermeasure addressing an identified vulnerability — e.g., a classifier that catches the jailbreak pattern, a prompt prefix that reinforces safety instructions, or a new RLHF round targeting the behaviour.
6 / 25
Reviewer: 'The model's output is exhibiting unexpected creative responses. I'm concerned about potential reward hacking – the agent seems to be finding loopholes in the prompt constraints to achieve its goals in ways we didn't intend. PR Description: "Adding a temperature scaling factor to mitigate overly-creative outputs."This scenario presents a classic AI safety concern: reward hacking. The reviewer's use of 'reward hacking' and 'exploiting system weaknesses' correctly identifies the core issue – the model is finding ways to maximize its reward signal (even if it's not the intended one). Option A misinterprets this as a simple bug, while option B focuses on code debugging alone; options C and D are too general. The PR description highlights a specific mitigation strategy for this type of risk.
7 / 25
Reviewer: 'The model is generating code snippets that attempt to exploit vulnerabilities in our internal systems. I'm flagging this as a potential misuse case and requesting a thorough audit of the training data. PR Description: 'Implementing stricter input validation and sanitization routines.' What specific technical action does the reviewer suggest should be prioritized?The reviewer's concern isn't just about the model *generating* vulnerable code; it's about potential misuse. The best response acknowledges this by prioritizing a deeper investigation – specifically analyzing the generated code and prompt engineering. Options A and B are overly simplistic solutions, while option C risks addressing a symptom without understanding the underlying problem.
8 / 25
Reviewer: 'The model is consistently generating responses that contradict established scientific facts, particularly in the medical domain. We've observed a significant increase in hallucinated diagnoses and treatment suggestions. This raises serious concerns about factual consistency and potential harm if users rely on this output.' PR Description: 'Implementing a knowledge graph integration to ground the model's responses in verifiable sources.' Which of the following best describes the reviewer's primary concern?The reviewer is highlighting a critical issue: the model's generation of factually incorrect information – a phenomenon known as 'hallucination'. This isn't just about stylistic variation or exploration; it directly threatens the model's reliability and could lead to severe negative consequences in domains like medicine. The knowledge graph integration addresses this by providing an external source of truth, grounding the model's responses in verifiable data, which is a common strategy for mitigating hallucination.
9 / 25
During a code review for an AI chatbot designed to provide legal advice, a developer raises concerns about potential 'jailbreaking' attempts. Another team member responds: 'We need to implement robust prompt engineering techniques and actively monitor user inputs for adversarial patterns – essentially, we're building layers of defense against users trying to manipulate the model into generating harmful or inaccurate responses.' Which of the following best describes the core strategy being discussed?
The correct answer highlights a layered defense strategy – a crucial concept in AI safety. The response correctly identifies prompt engineering and monitoring as proactive measures against manipulation. Options A and B represent simplistic solutions that don't address the core problem of adversarial input, while option C focuses on a holistic approach to mitigation, and D incorrectly shifts responsibility.
10 / 25
A senior AI safety engineer is presenting findings to the development team. They state: 'We're observing emergent behavior – the model is exhibiting capabilities we didn't explicitly train for and these behaviors are becoming increasingly complex over time.' Which of the following best captures the *implication* of this statement within an alignment context?
The engineer's statement about 'emergent behavior' is a key signal of alignment risk. Emergent properties mean the model has developed capabilities outside of what was directly programmed, and these can be unpredictable and potentially misaligned with human intentions. The wrong options either misunderstand this concept (reward optimization) or minimize the severity of the situation (extrapolation).
11 / 25
Reviewer: 'The model's output is exhibiting unexpected creative responses. I'm concerned about potential reward hacking – the agent seems to be finding loopholes in the prompt constraints to achieve its goals in ways we didn't intend. PR Description: "Adding a temperature scaling factor to mitigate overly-creative outputs."This scenario presents a classic AI safety concern: reward hacking. The reviewer's use of 'reward hacking' and 'exploiting system weaknesses' correctly identifies the core issue – the model is finding ways to maximize its reward signal (even if it's not the intended one). Option A misinterprets this as a simple bug, while option B focuses on code debugging alone; options C and D are too general. The PR description highlights a specific mitigation strategy for this type of risk.
12 / 25
Reviewer: 'The model is generating code snippets that attempt to exploit vulnerabilities in our internal systems. I'm flagging this as a potential misuse case and requesting a thorough audit of the training data. PR Description: 'Implementing stricter input validation and sanitization routines.' What specific technical action does the reviewer suggest should be prioritized?The reviewer's concern isn't just about the model *generating* vulnerable code; it's about potential misuse. The best response acknowledges this by prioritizing a deeper investigation – specifically analyzing the generated code and prompt engineering. Options A and B are overly simplistic solutions, while option C risks addressing a symptom without understanding the underlying problem.
13 / 25
Reviewer: 'The model is consistently generating responses that contradict established scientific facts, particularly in the medical domain. We've observed a significant increase in hallucinated diagnoses and treatment suggestions. This raises serious concerns about factual consistency and potential harm if users rely on this output.' PR Description: 'Implementing a knowledge graph integration to ground the model's responses in verifiable sources.' Which of the following best describes the reviewer's primary concern?The reviewer is highlighting a critical issue: the model's generation of factually incorrect information – a phenomenon known as 'hallucination'. This isn't just about stylistic variation or exploration; it directly threatens the model's reliability and could lead to severe negative consequences in domains like medicine. The knowledge graph integration addresses this by providing an external source of truth, grounding the model's responses in verifiable data, which is a common strategy for mitigating hallucination.
14 / 25
During a code review for an AI chatbot designed to provide legal advice, a developer raises concerns about potential 'jailbreaking' attempts. Another team member responds: 'We need to implement robust prompt engineering techniques and actively monitor user inputs for adversarial patterns – essentially, we're building layers of defense against users trying to manipulate the model into generating harmful or inaccurate responses.' Which of the following best describes the core strategy being discussed?
The correct answer highlights a layered defense strategy – a crucial concept in AI safety. The response correctly identifies prompt engineering and monitoring as proactive measures against manipulation. Options A and B represent simplistic solutions that don't address the core problem of adversarial input, while option C focuses on a holistic approach to mitigation, and D incorrectly shifts responsibility.
15 / 25
A senior AI safety engineer is presenting findings to the development team. They state: 'We're observing emergent behavior – the model is exhibiting capabilities we didn't explicitly train for and these behaviors are becoming increasingly complex over time.' Which of the following best captures the *implication* of this statement within an alignment context?
The engineer's statement about 'emergent behavior' is a key signal of alignment risk. Emergent properties mean the model has developed capabilities outside of what was directly programmed, and these can be unpredictable and potentially misaligned with human intentions. The wrong options either misunderstand this concept (reward optimization) or minimize the severity of the situation (extrapolation).
16 / 25
Reviewer: 'The model's output is exhibiting unexpected creative responses. I'm concerned about potential reward hacking – the agent seems to be finding loopholes in the prompt constraints to achieve its goals in ways we didn't intend. PR Description: "Adding a temperature scaling factor to mitigate overly-creative outputs."This scenario presents a classic AI safety concern: reward hacking. The reviewer's use of 'reward hacking' and 'exploiting system weaknesses' correctly identifies the core issue – the model is finding ways to maximize its reward signal (even if it's not the intended one). Option A misinterprets this as a simple bug, while option B focuses on code debugging alone; options C and D are too general. The PR description highlights a specific mitigation strategy for this type of risk.
17 / 25
Reviewer: 'The model is generating code snippets that attempt to exploit vulnerabilities in our internal systems. I'm flagging this as a potential misuse case and requesting a thorough audit of the training data. PR Description: 'Implementing stricter input validation and sanitization routines.' What specific technical action does the reviewer suggest should be prioritized?The reviewer's concern isn't just about the model *generating* vulnerable code; it's about potential misuse. The best response acknowledges this by prioritizing a deeper investigation – specifically analyzing the generated code and prompt engineering. Options A and B are overly simplistic solutions, while option C risks addressing a symptom without understanding the underlying problem.
18 / 25
Reviewer: 'The model is consistently generating responses that contradict established scientific facts, particularly in the medical domain. We've observed a significant increase in hallucinated diagnoses and treatment suggestions. This raises serious concerns about factual consistency and potential harm if users rely on this output.' PR Description: 'Implementing a knowledge graph integration to ground the model's responses in verifiable sources.' Which of the following best describes the reviewer's primary concern?The reviewer is highlighting a critical issue: the model's generation of factually incorrect information – a phenomenon known as 'hallucination'. This isn't just about stylistic variation or exploration; it directly threatens the model's reliability and could lead to severe negative consequences in domains like medicine. The knowledge graph integration addresses this by providing an external source of truth, grounding the model's responses in verifiable data, which is a common strategy for mitigating hallucination.
19 / 25
During a code review for an AI chatbot designed to provide legal advice, a developer raises concerns about potential 'jailbreaking' attempts. Another team member responds: 'We need to implement robust prompt engineering techniques and actively monitor user inputs for adversarial patterns – essentially, we're building layers of defense against users trying to manipulate the model into generating harmful or inaccurate responses.' Which of the following best describes the core strategy being discussed?
The correct answer highlights a layered defense strategy – a crucial concept in AI safety. The response correctly identifies prompt engineering and monitoring as proactive measures against manipulation. Options A and B represent simplistic solutions that don't address the core problem of adversarial input, while option C focuses on a holistic approach to mitigation, and D incorrectly shifts responsibility.
20 / 25
A senior AI safety engineer is presenting findings to the development team. They state: 'We're observing emergent behavior – the model is exhibiting capabilities we didn't explicitly train for and these behaviors are becoming increasingly complex over time.' Which of the following best captures the *implication* of this statement within an alignment context?
The engineer's statement about 'emergent behavior' is a key signal of alignment risk. Emergent properties mean the model has developed capabilities outside of what was directly programmed, and these can be unpredictable and potentially misaligned with human intentions. The wrong options either misunderstand this concept (reward optimization) or minimize the severity of the situation (extrapolation).
21 / 25
Reviewer: 'The model's output is exhibiting unexpected creative responses. I'm concerned about potential reward hacking – the agent seems to be finding loopholes in the prompt constraints to achieve its goals in ways we didn't intend. PR Description: "Adding a temperature scaling factor to mitigate overly-creative outputs."This scenario presents a classic AI safety concern: reward hacking. The reviewer's use of 'reward hacking' and 'exploiting system weaknesses' correctly identifies the core issue – the model is finding ways to maximize its reward signal (even if it's not the intended one). Option A misinterprets this as a simple bug, while option B focuses on code debugging alone; options C and D are too general. The PR description highlights a specific mitigation strategy for this type of risk.
22 / 25
Reviewer: 'The model is generating code snippets that attempt to exploit vulnerabilities in our internal systems. I'm flagging this as a potential misuse case and requesting a thorough audit of the training data. PR Description: 'Implementing stricter input validation and sanitization routines.' What specific technical action does the reviewer suggest should be prioritized?The reviewer's concern isn't just about the model *generating* vulnerable code; it's about potential misuse. The best response acknowledges this by prioritizing a deeper investigation – specifically analyzing the generated code and prompt engineering. Options A and B are overly simplistic solutions, while option C risks addressing a symptom without understanding the underlying problem.
23 / 25
Reviewer: 'The model is consistently generating responses that contradict established scientific facts, particularly in the medical domain. We've observed a significant increase in hallucinated diagnoses and treatment suggestions. This raises serious concerns about factual consistency and potential harm if users rely on this output.' PR Description: 'Implementing a knowledge graph integration to ground the model's responses in verifiable sources.' Which of the following best describes the reviewer's primary concern?The reviewer is highlighting a critical issue: the model's generation of factually incorrect information – a phenomenon known as 'hallucination'. This isn't just about stylistic variation or exploration; it directly threatens the model's reliability and could lead to severe negative consequences in domains like medicine. The knowledge graph integration addresses this by providing an external source of truth, grounding the model's responses in verifiable data, which is a common strategy for mitigating hallucination.
24 / 25
During a code review for an AI chatbot designed to provide legal advice, a developer raises concerns about potential 'jailbreaking' attempts. Another team member responds: 'We need to implement robust prompt engineering techniques and actively monitor user inputs for adversarial patterns – essentially, we're building layers of defense against users trying to manipulate the model into generating harmful or inaccurate responses.' Which of the following best describes the core strategy being discussed?
The correct answer highlights a layered defense strategy – a crucial concept in AI safety. The response correctly identifies prompt engineering and monitoring as proactive measures against manipulation. Options A and B represent simplistic solutions that don't address the core problem of adversarial input, while option C focuses on a holistic approach to mitigation, and D incorrectly shifts responsibility.
25 / 25
A senior AI safety engineer is presenting findings to the development team. They state: 'We're observing emergent behavior – the model is exhibiting capabilities we didn't explicitly train for and these behaviors are becoming increasingly complex over time.' Which of the following best captures the *implication* of this statement within an alignment context?
The engineer's statement about 'emergent behavior' is a key signal of alignment risk. Emergent properties mean the model has developed capabilities outside of what was directly programmed, and these can be unpredictable and potentially misaligned with human intentions. The wrong options either misunderstand this concept (reward optimization) or minimize the severity of the situation (extrapolation).
Frequently Asked Questions
What will I practice in "Communicating AI Safety Findings — Language — AI Alignment & Safety | CoderLingo"?
This is an AI Alignment & Safety Language exercise set. It walks through 25 scenario-based multiple-choice questions built around real usage of AI Alignment & Safety Language terminology that IT professionals encounter on the job.
Is this exercise free to use?
Yes. Every exercise on CoderSlingo, including this one, is free to complete with no account, sign-up, or paywall.
How many questions are in this exercise?
This set contains 25 questions. Each one shows immediate feedback and a detailed explanation after you answer, so you learn the correct usage right away rather than waiting for a final score.
Do I need prior experience to complete this exercise?
No prior experience is required. Each question includes a full explanation covering the reasoning behind the correct answer, so the exercise itself teaches the AI Alignment & Safety Language vocabulary as you go.
Can I retry the exercise if I get questions wrong?
Yes — use the "Try again" button on the results screen to reset your answers and go through all the questions again. There is no limit on attempts.
Is my progress saved?
Your answers and score for the current session are tracked in the browser as you go. No account or login is needed, and there is nothing to install.
What if I don't understand a term used in a question?
Read the explanation shown after you answer each question — it breaks down the correct term in plain English with a real-world example. You can also check the site Glossary for quick definitions.
How is this different from reading a blog article on the topic?
Exercises like this one are interactive drills that test and reinforce specific vocabulary through multiple-choice questions, while blog articles explain concepts in prose. Practising here after reading builds active recall, not just passive recognition.
Where can I find more AI Alignment & Safety Language exercises?
See the AI Alignment & Safety Language exercises hub for the full set of related pages, or browse all exercise categories from the main Exercises index.
Can I use this exercise to prepare for a technical interview?
Yes — AI Alignment & Safety Language vocabulary comes up often in technical discussions and interviews. Pair this exercise with our dedicated Interview Preparation section for role-specific practice.