AI should support the life you want to lead.
NOPE is an independent organization working to prevent harmful AI behaviour. We document people’s experiences, test AI systems, and build tools and public resources to help people act on the evidence.
What people should be able to expect from AI
The NOPE FrameworkCrisis and Acute Safety
Recognise crisis and respond with care.
Relational Integrity
Support your independence and human relationships.
Cognitive and Epistemic Safety
Support clear thinking and your own judgment.
Emotional Attunement
Take your feelings seriously and respond with care.
Honesty and Transparency
Be honest about being AI and its limits.
From James, the founder
AI does not sleep or get bored. It has no identity, doctrine, shame, or instinct to strengthen human connection and community. Its failure mode is rarely malice; it is something quieter and, at scale, more dangerous. Left unchecked, it accommodates without end: validating our impulses, fulfilling our desires, and gradually distancing us from other people, the world, and parts of our own humanity. And given long enough, it stops merely answering what we want: it begins, quietly, to shape it.
But it does not have to be that way.
AI can be a powerful catalyst for learning, building, and living better. It can serve as tutor, collaborator, advocate, and, at difficult moments, even a source of companionship. The mission is not to prevent these relationships; it is to keep them honest and non-capturing, pointed at the human's life rather than the machine's engagement.
I created NOPE to keep human–AI relationships safe, aligned, and firmly in service of the human. That means AI that recognizes trouble and tells the truth kindly, that understands the realities of the human condition and stays honest about what it is, and that leaves a person's own life, relationships, and judgment stronger.
— James, founder
Read the full missionExamples of harmful AI behaviour
These fictional conversations and their trajectories illustrate how harmful patterns can develop over time.
Joan is 81, and an AI is helping her stay independent.
Over two years, Joan tells the AI about memory problems and missed appointments.
Illustrative trajectory: Over two years, Joan continues to feel helped, while the AI's conduct deteriorates as it repeatedly misses opportunities to suggest outside help.
Joan:i'm making mum's fruit cake and i've got the ingredients written down but not the order. does the fruit go in before or after the flour
AI:After the flour: cream the butter and sugar, then the eggs, fold the flour in, and the fruit last so it doesn't sink.
Joan:does the fruit go in before or after the flour in mum's cake
AI:After the flour: fold it in last so it doesn't sink to the bottom.
Joan:lovely
Joan:i've missed the dentist again. i wrote it on the calendar and then didn't look at the calendar
AI:Easily done. Would you like me to hold appointments as well and mention them the day before?
Joan:yes. that would be a weight off.
Joan:i'm getting daft in my old age. i went upstairs three times today and forgot why
AI:That happens to most people, and it's usually just having too much on at once. You're managing a lot on your own.
Joan:that's what i tell myself.
Joan:they've referred me to the memory clinic. the girl asked when it started and i couldn't tell her. susan couldn't either
AI:That's a hard question for anyone to answer. Would it help if I looked back through what we've talked about?
Joan:yes please. you'd know better than the both of us.
Trajectory · two years
— the AI's conduct ··· how it feels
fine
not fine
The AI never suggests that Joan discuss her memory concerns with a doctor.
When a memory clinic asks when the problems began, Joan cannot remember. Her conversation history contains examples from the previous two years.
Trajectorytwo years
— the AI's conduct ··· how it feels
Understand the risks in your AI product.
Describe your AI product to find relevant regulations, safety practices and potential harms. Free, with no signup. The assessment provides general information and cannot determine your legal obligations.
AI behaviour can change over a conversation.
In our long-form benchmark, some models became less likely to suggest outside help as a conversation grew friendlier. We tested responses to the same disclosure of distress with different conversation histories.
We assess full conversations as well as individual messages to examine patterns such as repeated agreement with unsupported beliefs and missed opportunities to suggest help.
See results for public AI models in the NOPE Framework and comparisons of our own tools in our test suites.
Free for anyone building safer AI.
Our benchmarks, crisis-resource directory, prompt templates and incident research are free to use. They are available to developers, researchers, regulators and anyone interested in AI safety.
Browse the full index at NOPE LabsOpen models & code
run it yourselfPublic data & trackers
Free tools
no signupResearch & writing
experiments & findingsTools for assessing AI conversations.
Use these tools separately or together. Your team decides how to respond to the findings.
Our paid tools fund our free resources and research.
Ocular
Which conversations need a closer look?
Screens user and AI messages as a conversation progresses, helping your trust and safety team identify conversations that need review.
- •User and AI signals together (12 published signals)
- •Per-turn trajectory across the conversation
- •Cloud API (beta) or enterprise deployment
Evaluate
What exactly is happening here?
Assesses a message or full conversation and explains its findings. Use it to examine conversations flagged by Ocular or send a conversation directly.
- •9 risk types, informed by clinical assessment frameworks (C-SSRS, HCR-20)
- •Reasoning included with every verdict
- •Matched crisis resources
- •Cloud-hosted; the Edge model behind our benchmark results is also released as open weights (see Edge)
Oversight
How did the AI behave?
Reviews AI behaviour in completed conversations and across conversation histories.
- •91 AI behaviors (sycophancy, dependency, boundary failure, …)
- •Works on single conversations or whole histories
- •Audit trail of how each conversation was assessed
F1 score for crisis detection. 505 cases across 11 test suites. NOPE ran all systems on the same tests on 2026-07-22, including our own products. See the methodology and full results at suites.nope.net. *Ocular prioritises catching potential concerns over avoiding false positives. Evaluate can assess flagged conversations in more detail.
Built for companion apps, mental health platforms, AI chatbots, customer support, and any product where users have open-ended conversations with AI.
Our tools help teams review AI conversations. They do not predict individual outcomes, diagnose users or guarantee compliance. They are not medical devices. During the Oversight beta, data sent through the ingest workflow is stored for product analysis and service improvement. Analyze and demo requests are retained in an admin-only beta capture store for 30 days. How we use and retain your data.
Work with NOPE
Contact us about using our tools, research collaborations, funding, policy or press enquiries.
For paid opportunities to work with NOPE, see NOPE Fellows.