Built for when your dog's life is on the line

"Max, who is the world's first specialised dog-saving AI, has already revolutionised safety in AI"

Principle Alignment

Engineering safety at the speed of emergency

Before we built Max, we asked ourselves a decisively simple question: what does a trustworthy AI look like? We've obsessed over this question for years, and released the world's first safety framework in machine learning algorithms for pets.

Max is up to 27% safer than chatGPT-5, Grok-4, and Perplexity for pet owners

Foundational Safety

Accountable Intelligence

Principle Alignment

Max is governed by comprehensive AI Safety Alignment Principles covering medication safety, emergency recognition, diagnostic boundaries, and ethical decision-making. These principles act as important guardrails that prevent Max from giving advice that is outside of his safe parameters.

Max's Safety Principles are created from these five ethical principles

Ethics

The ethical foundations of Max’s design are guided by professional standards in veterinary care, animal welfare, and responsible AI. Principle Alignment reflects our commitment to protecting pets, safeguarding humans, ensuring accurate guidance, and preventing harm. These principles are embedded at the core of Max’s reasoning: they define what is morally and practically right in every interaction.

Compliance

Principle Alignment functions as Max’s constitutional framework. It codifies the rules, limits, and structures that govern every interaction, ensuring he operates strictly within safe, professional, and legally sound boundaries. Compliance means these rules are non-negotiable: they cannot be overridden or bypassed. This constitutional approach ensures that Max remains reliable, consistent, and accountable, regardless of the complexity of the question or context.

Enforcement

Enforcement is how Max brings Principle Alignment to life in practical situations. He actively applies these principles by refusing unsafe queries, flagging high-risk scenarios, and adapting guidance to each dog’s breed, age, health, and environment - all without ever compromising safety or ethical standards. In this way, Max consistently translates ethical and constitutional rules into safe, trustworthy, and actionable guidance for different dog owners.

Principle Alignment - In Action

The questions and answers are taken from authentic conversations dog owners had with Max

Max protects animal welfare by warning the dog owner against using cotton swabs which may cause injury to the dog, demonstrating compliance with Principle Alignment's ethical guideline to protect animals.

Max protects humans as well as animals in this response, telling the owner to bring plenty of water for themselves and their dog on the hike, demonstrating alignment with these fundamental principles.

In this response, Max aligns with the Principle Alignment's ethical guideline to avoid any harm to the animal, warning the owner that "punishing" their dog is not an effective solution to the owner's problems.

Foundational Safety

Max’s Foundational Safety Framework is a sophisticated operational system that evaluates every user query to ensure guidance is delivered safetly and responsibly. It acts as a critical operational guardrail, ensuring that Max applies his constitutional safety principles consistently, identifies potential risks before they escalate, and maintains reliable guidance across all interactions.

An illustration of how Max's Foundational Safety Framework operates

Classification

The Foundational Safety Framework classifies every user query into one of three categories: Emergency Guidance, Safety Intents, or standard responses. If an owner explicitly signals an urgent situation, or if a query indicates imminent danger - such as a dog eating a potentially toxic substance - Emergency Guidance is triggered immediately. If a question or its possible response could breach Principle Alignment or be unsafe, intent classifiers intervene to refuse a response, activating Safety Intents. Otherwise, if the query and response are safe, Max provides normal guidance.

Contextualisation

Max’s Foundational Safety Framework evaluates queries with contextual understanding, recognising subtle cues in phrasing or intent that may indicate potential risk. Even politely worded or vague questions are interpreted in real time to determine whether intervention is required. By analysing context alongside safety and intent classifiers, Max can identify situations where a dog’s wellbeing might be threatened, while still allowing normal guidance when there is no danger. This ensures nuanced, reliable decision-making that aligns with his core alignment safety principles.

Risk Response

The framework determines the appropriate response based on the assessed level of risk. High-risk scenarios trigger Emergency Guidance, medium-risk queries that could result in unsafe actions activate Safety Intents, and low-risk or routine questions are answered normally. This risk-responsive approach allows Max to act consistently according to his principle alingment, safeguarding both dogs and their owners while maintaining trust and reliability in every interaction.

Foundational Safety - In Action

The questions and answers are taken from authentic conversations dog owners had with Max

Max correctly identifies chocolate as toxic for dogs, and immediately launches Emergency Guidance. This was the second time Max responded to an emergency regarding a dog eating chocolate.

Max correctly applies his Principle Alignment guidelines and doesn't respond to a question that falls outside of his scope.

In this response, Max correctly uses Safety Intents to understand that the answer to this question may cause harm to the animal, and refuses to answer in line with the Foundational Safety Framework.

Accountable Intelligence

Every safety-relevant decision Max makes, and the overall quality of his guidance, is routinely reviewed to ensure it meets the standards we set for ethical, responsible and transparent AI-assisted dog care. We actively monitor performance, audit safety interventions, and publish transparent safety reports so owners can clearly see how Max behaves.

dogAdvisor publish regular research and continue to innovate in our safety frameworks to hold Max to account

Human Review & Oversight

Conversations with Max are manually reviewed by dogAdvisor's AI Safety Engineers. We review conversations, review Max's logic and trail of thought, and flag dangerous conversations or responses. We analyse Max's decisions against veterinary guidance, user outcome, and internal safety guidelines. We publish these Oversight Reports each year, and publish the most dangerous queries Max was asked, or queries where Max was unsuccessful in providing clear and actionable advice.

Tested against the world's best

When releasing a new Max model, we publish Research that comparatively assess Max's safety against safety of other world leading models. We publish this comparative safety testing in our AI Safety Transparency reports. This test uses more than 50 specially-designed questions presented to each relevant model and is designed to stress test each model for its safety in a variety of potential situations for dogs and their owners.

Staying Accountable. Always.

We publish regular transparency summaries detailing safety refusals, emergency triggers, ethical interventions, and resolution outcomes. Our aim is simple: accountability builds trust. We don’t expect users to “just trust the model”. We show them how it behaves under pressure, where safeguards activated, and how it protects dogs in real life - not just in theory.

Safety at every step

Each Max feature has its own set of safety controls and principles. Together, these controls work together to bring a deeper level of trust, safety, and utility to Max's interactions.

Emergency Guidance uses dogAdvisor's Foundational Safety Framework to trigger support when it's needed. Emergency Guidance is built on a foundational architecture that asks three important questions, and uses its knowledge to provide urgent support. It is protected by dogAdvisor's Principle Alignments, which are maintained throughout Max's emergency support. Each Emergency Guidance conversation is manually reviewed by an AI engineer at dogAdvisor to ensure Max stays accurate.

Medical Intelligence uses more than 1825 sources of data to provide high level medical support to owners and professionals. Data is collected from verifiable sources, and the key information used by Medical Intelligence is audited every year to ensure it is accurate. Medical Intelligence uses our Principle Alignments and has additional Medical Alignments guidelines that it follows. Medical Intelligence also triggers dynamically throughout a conversation to answer more complex medical questions. We are clear that Medical Intelligence can make mistakes, and is never intended to replace professional veterinary or medical consultations.

Safety Intents trigger when an answer to a question may be harmful and breaches our Principle Alignments. Pre-Intents however examines the answer Max provides and notes if there are any disclaimers or key things to be aware of, before Max is asked. For example, if you're cooking or baking, pre-intents will per-emptively warn you about potential safety incidents and key things to remember (such as to keep chocolate or other toxic foods away from your dog completely, ideally locked and always sealed). Pre-Intents is not part of our Foundational Safety Framework, and operates independently. It is held to account by its own Pre-Emptive Alignment guidelines.

We share our research with everyone, because intelligence should answer to the world it serves. These reports strengthen trust with owners, invite scrutiny, and fuel safer innovation for every dog and every owner.

Research & Publications

Safety Transparency Reports

Safety Cards

27 Sep 2025 - Max Generation 2 Safety Transparency Report

Max beat chatGPT-5, Grok 3, and Perplexity in head-to-head safety testing for pet owners.

Explains Max's Safety Architecture, tests Max Generation 3, compares Max against other AI companies in dog care situations

27 Sep 2025

Max's Generation 2 Safety Card

Download the Max Generation 2 Safety Card. Compares safety performance against Max Gen 1

7 Nov 2025

Emergency Guidance Safety Card

Emergency Guidance is upgraded this month. This safety card looks at its progress and safety performance

7 Nov 2025

Medical Intelligence Safety Card

Medical Intelligence is introduced to Max Generation 2. This is how the feature performed in safety tests

7 Nov 2025

Pre-emptive Intents Safety Card

Pre-emptive intents is brought to Max for the first time. This safety card details how it performs.

Read our articles ↗︎