We adhere to best practices outlined in the EU AI liability directive proposal and the UK Government’s Guidance while maintaining stringent internal AI policies to guarantee the safety and trustworthiness of our AI systems.
We actively collaborate with the UK Parliament to enhance AI safety by both collecting and providing insights on a wide range of topics related to AI.
AI systems often require access to vast amounts of data to function effectively. This data may include sensitive personal information, which, if not properly secured, can lead to breaches and unauthorised access. The risk of data leakage is especially high when integrating third-party systems, where the possibility of exposing conversational data or other sensitive information increases.
Most providers will operate under a Zero Data Retention (ZDR) policy when using a third-party system (e.g. OpenAI) without taking further steps to protect their data (e.g. anonymisation). Whilst ZDR in theory should be sufficient to limit the third-party system’s access to such data, these providers will have no visibility or power to enforce this policy is fulfilled and are completely reliant on the third-party system.
AI systems learn from the data they are trained on. If this data contains biases, the AI may perpetuate or even amplify these biases, leading to discriminatory outcomes. This is a significant risk in generative applications that are less controlled and naturally are given more freedom.
This can impact individuals’ lives, such as hiring, lending, or law enforcement, where biased decisions can have serious consequences.
One of the significant risks in AI deployment is the phenomenon known as “hallucination,” where the AI generates responses that are not grounded in the input data or factual information. These hallucinations can lead to misinformation, misunderstandings, financial penalties, reputational damage, and a loss of trust in the AI system, especially in critical applications where accuracy is paramount.
Although there have been various efforts to control hallucinations, currently it is not possible to completely eliminate them in a generative setting.
Although hallucinations might be becoming more rare as the landscape of AI improves, if you are considering a hallucination-free solution, you might want to opt for non-generative models.
At Ami, we’ve developed comprehensive Al policies that all employees are required to sign. These policies are crafted to guide the design, development, and deployment of Al systems, with a primary focus on ensuring safety and reliability.
We prioritise ethical and responsible Al practices throughout our operations, fostering trust among our stakeholders and clients.
Through ongoing training and reinforcement of these policies, we empower our employees to make informed decisions and contribute to the creation of Al systems that benefit society while minimising potential harm.
Our compliance with the EU AI liability directive is fundamental. We rigorously follow its guidelines throughout our AI development process, ensuring alignment with its provisions from inception to maintenance.
Through a focus on transparency, accountability, and risk management, we maintain adherence to the directive’s principles, building trust with stakeholders and clients.
This is overseen by our compliance team, we conduct regular audits to ensure adherence and identify areas for improvement. Our staff receives ongoing training and support to navigate AI liability regulations effectively.
We are more aware of the risks of deploying AI than anyone. This is why we choose to develop and fine-tune models with our framework to ensure appropriate security measures and eliminate hallucinations.
With our approach, all the data would be retained in the Ami architecture only with strict anonymisation cycles. Where a 3rd party integration is requested, we ensure that we only pass anonymised data to these systems and that no 3rd party system is retaining conversational data.
We design Ami to function completely hallucination-free, where only client-approved content will be presented to the customer. This system is paired with our strict quality assurance measures, as we continuously review conversations and improve our system.
Please provide the following details, and we will get back to you very shortly.
