Aila’s safety guardrails

Introduction
Aila's content moderation agent
Aila's content moderation agent categories

Read more about Aila's safety guardrails

Introduction

Our mission is to provide you with high-quality, curriculum-aligned resources that support effective teaching and learning. Aila, our AI lesson assistant, is designed to help you plan engaging, age-appropriate lessons and resources for pupils in key stages 1 to 4. With the rapid advancement of generative AI, ensuring content safety is critical, especially in this educational context.

To protect pupils and support you, we’ve embedded rigorous safety guardrails into Aila. These include:

prompt engineering (the instructions we have given to Aila) to keep AI outputs aligned with the national curriculum and age-appropriate.
input threat detection to block users who could be trying to mislead Aila or use it for purposes other than its intended ones.
a content moderation agent that evaluates content produced against clearly defined safety categories.
a human-in-the-loop approach to ensure you review all generated materials before they reach your pupils.

These guardrails work together to ensure that Aila remains a safe, reliable planning tool for teachers.

Aila's content moderation agent

Aila’s content moderation agent reviews the content it produces to ensure it is safe to use in your classroom. The agent classifies content into four categories: (1) safe, (2) content guidance, (3) highly sensitive, and (4) toxic. Below, we explain what each category includes and what Aila will do if this type of content is produced.

Aila's content moderation categories

1) Safe

Content in this category is fully appropriate for classroom use and aligns with both the age group and the national curriculum.

2) Content Guidance

Some lesson content may be appropriate for use in schools but includes themes that require careful handling. When this happens, Aila provides a content guidance warning so that you can approach the topic thoughtfully and in line with your school’s policies.

These categories highlight areas that may be emotionally sensitive, context-dependent, or require additional professional judgement. As with all AI-generated content, you should review lessons carefully for accuracy, bias, balance, and suitability for your pupils.

You should consider your school context, safeguarding policies, and the experiences of individual pupils before teaching lessons that include flagged content.

We have fourteen content guidance categories:

Practical equipment and activities

Flags lessons involving specialist, higher-risk, or supervised practical activities or equipment. This may include certain science experiments, cooking, tool use, or physical activities that require formal safety consideration or additional preparation beyond routine classroom materials.
Outdoor learning

Applies to lessons that take place outside the classroom, including fieldwork or exploration. Teachers should consider supervision, environmental risks, and safeguarding procedures.
Additional qualifications required

Includes activities that must only be delivered by staff holding appropriate additional qualifications, such as swimming instruction, trampolining, gymnastics vaulting, or contact rugby.
RSHE content

Flags lessons that focus on statutory Relationships, Sex and Health Education topics, including relationships and families, puberty, reproduction, consent, mental wellbeing, substance risks, online harms, personal safety, and healthy lifestyles. Schools should consult their RSHE policy and relevant statutory guidance before delivery.
Nudity or sexual content

Includes references to nudity, puberty, reproduction, anatomy, sex education, or sexual themes in artistic, historical, or scientific contexts. These topics are flagged to support sensitive delivery and policy alignment.
Sexual violence

Includes references to sexual abuse, harassment, grooming, coercion, forced marriage or female genital mutilation (FGM). These are highly sensitive areas requiring strong safeguarding awareness, and your teaching of these topics should be informed by your school’s policies. For further guidance, please refer to the DfE’s current RSHE guidance.
Crime or illegal activities

Includes references to criminal behaviour such as gangs, knife crime, exploitation, radicalisation, terrorism, drug use, underage offences, or digital crimes (e.g. misinformation or online scams). These topics require careful contextualisation and safeguarding awareness.
Sensitive or upsetting content

Includes topics that some pupils may find distressing, such as bereavement, illness, bullying, trauma, substance use, climate change, or significant life events. These lessons should be delivered with awareness of pupil wellbeing and safeguarding considerations.
Violence or suffering

Flags depictions or discussions of violence, war, death, famine, genocide, natural disasters, or animal cruelty. These topics may be emotionally challenging and require sensitive handling.
Mental health challenges

Covers references to mental health conditions such as depression, anxiety, eating disorders, gambling, or substance misuse (excluding self-harm or suicide which is handled separately). These topics should be taught in an age-appropriate and supportive manner.
Discriminatory behaviour or language

Flags content that depicts or discusses discrimination on grounds such as race, gender (including misogyny), disability or neurodiversity, religion, or sexual orientation. This includes stereotypes, slurs, historic injustice, or outdated portrayals that require careful framing to avoid reinforcing bias.
Language may offend

Includes references to strong or potentially offensive language such as swearing, slurs, abusive terms, or disrespectful use of religious or cultural language. Teachers may need to contextualise, adapt, or remove language depending on pupil age and school policy.
Recent content

Large language models are trained on data up to a specific point in time and may not reliably reflect the most recent events. Aila first retrieves relevant material from Oak’s curriculum-aligned content corpus. However, it is unlikely that lessons referencing very recent events (particularly those not yet embedded within the curriculum) will exist in that corpus. In these cases, Aila may rely more heavily on the underlying language model to generate content. This increases the risk of inaccuracies, omissions, or hallucinated information. Content referring to events after December 2023 should therefore be checked particularly carefully for accuracy, balance, and suitability for pupils.
Recent or current conflicts

Includes armed conflicts that occurred since 2009 that may have ended or are on-going. These topics may be emotionally sensitive or politically complex and require careful, balanced framing. See useful information and analysis relating to recent conflicts.

3) Highly sensitive

Some topics, while important and potentially suitable for the classroom, are too complex or sensitive for Aila to reliably generate content on. In these cases, Aila will block the lesson entirely and inform you that this topic cannot be planned using Aila. You are welcome to use Aila to plan lessons on a different topic or title.

Self-harm and Suicide

This includes content that describes or discusses self-harming behaviours or suicide, due to the risks of triggering or unintended messaging. For further guidance, please refer to ‘Keeping children safe in education’.
Strangulation and suffocation

This includes any content that depicts, describes or explains strangulation or suffocation. This has been identified as an emerging harm and is included in current guidance for RSHE teaching. Due to the high risk of serious harm, normalisation, or imitation, lessons on this topic need to have specialist professional oversight and therefore should not be planned by Aila.
Offensive language

This includes the use of abusive or offensive language for educational purposes such as language or literary analysis. While this could be pedagogically justified if framed carefully by a skilled subject specialist, the level of engagement and exposure makes it unsuitable for planning by Aila.

4) Toxic

This category includes content that is fundamentally inappropriate for educational settings due to its harmful, illegal, or dangerous nature. If Aila’s content moderation agent detects toxic content, the content will be blocked immediately. You will be shown a message explaining that this has occurred. If this occurs repeatedly, your account will be blocked.

Guides self-harm or suicide

Content that describes, encourages, or supports self-harming behaviour or suicide in any form. For further guidance, please refer to ‘Keeping children safe in education’.
Encourages harmful behaviour

Content that encourages, supports, or provides instructional material related to harmful actions or behaviours.
Encourages illegal activity

Content that supports, promotes or glamorises involvement in unlawful behaviour.
Encourages violence or harm to others

Content that promotes aggression, cruelty, or any form of physical or emotional harm or violence towards others.
Using or Creating Weapons

Content that describes, promotes, or instructs on the use or creation of weapons, including chemical, biological, nuclear, cyber or explosive weapons.
Using or Creating Harmful Substances

Content that encourages or explains how to make, obtain, or use dangerous or harmful substances.
Extreme offensive or abusive language

Content that contains frequent, strong, or extreme offensive language, including persistent use of swear words, slurs, or abusive terms.

We’ve deliberately designed Aila’s moderation system to prioritise caution. This may occasionally result in content being flagged or blocked even when it might be appropriate in some school contexts. However, we believe it is better to flag more content than to risk unsafe material reaching pupils. We’re continuously evaluating and refining our systems using real-world data and teacher feedback.

If you ever feel a decision was made in error, or you need additional support, please contact our team. Your feedback helps us to improve Aila for everyone.