Transcription
All right, hello everybody and welcome to this webinar on AI and the law, a 2025 guide. And the focus is on what every US attorney must know. And my name is Tolulope Awolowo Yemi and this is brought to you by AI and Law Breach.
Well, a little about myself, I am an AI educator, researcher and an an AI legal strategist. I have four years teaching AI to professionals. I'm a lawyer also and I am the author of breaching AI and law and the simple book of AI series. You can find them on Amazon and also courses on YouTube and Udemy. I'm also an academic. I have an LLM. I have a masters in information systems technology and AI and a professional certificate in AI and machine learning from MIT.
Now, what does this session, this webinar promise? First, it's going to be in clear language, no hype. It's real, explained clearly and succinctly. I'm going to also give you checklist and workflows that you can start to implement as soon as tomorrow. And also give you concrete risk controls mapped to the NIST, the NIST rules.
Now, what are you going to learn today? I'm going to explain core AI terms. We're also going to talk about the ABA Formal Opinion 512. And also some state bar guidance, Florida and New York. And we're also going to have a checklist. I'm going to give you 10 checklists for evaluating your vendors, your AI vendors before choosing them. And we're going to spot critical IT pitfalls and also and how you can advise clients on becoming EUAI timeline. Of course, some of some have already gone into implementation, but we'll discuss more on that later. Let's go.
Why does AI matter in 2025? A trifecta of reasons. First is industry pressure. Now, AI adoption is nearly tripled. So, I'm going to show you a graph. A chart has tripled from 11% in 2023 to 30% in 2024. Of course, I can bet that it's going to even quadruple or triple in 2025. And this is of course from ABA 2024 Legal Technology Survey.
Another reason is client's expectations. So, especially with corporate clients, they expect more use of Gen AI. Of course, and they're expecting that the prices or the costs would be reduced because it might be maybe less um work hours and so on. But we're going to talk more of that later on. So, because in-house leaders increasingly approve of law firm Gen AI use. So, this has invariably led to corporate clients especially expecting or anticipating reduced billings. And of course, more efficiencies with AI use.
And the third is bar guidance. State bars emphasize the duty of tech competence and supervision. The language is clear. Use AI, but verify and protect confidentiality.
So now, let's Let me explain some core AI concepts for lawyers. First is LLMs. I'm sure uh initially when I first ever heard this word some years ago, I thought, okay, so it doesn't mean masters of laws. So, yeah, LLMs means large language model. So, I really hear, think of it as an advanced auto complete. It's trying to predict the next word. It doesn't actually understand the context. It's just a machine that has learned or crammed the language so much that it's able to give you the next word and just conjure some group of words or elements in ways that make sense to you, but it actually doesn't understand what it it means. So, an analogy is an associate who has read every case, but doesn't grasp the core legal principles. That's what an LLM is. If an LLM were to be a human being, that's that's the kind of associates the LLM would be.
Now, hallucinations, yes, this is another buzzword. This is when AI confidently invents false information. Well, this is not a bug. This is an inherent feature of AI. An analogy to explain this is a junior associate inventing a plausible but fake case citations. The rule is verify, never trust. Never ever ever ever just take out AI's output and just use without verification. No. No.
So, the next is retrieval augmented generation, RAG for short. So, basically, it's Let me read it. The key safeguard that grounds an AI in trusted sources, forcing it to cite its work. So, you, for example, you your law firm wants to build their own internal like AI library. So, you first of all leverage large language models that have been trained with thousands of documents. But then you now use your own firm's research, your own firm's document. That's called RAG. So, that whatever it's outputting such as the citations, the cases, the statutes, it is referencing it to your own documents. So, that's what RAG is. So, an analogy is an open book exam where you hand an associate the specific case file and your firm's private research. So, basically, you're giving the AI, this is the answers. This is where I want you to get the answers from. Reference it from this document or this file that I'm giving you.
And the last is embeddings. Now, this is tech that finds information by conceptual meanings, not just keywords. So, um I I don't know if you used tools that you could when you search, it's searching only the word you put. But embeddings help the AI to not just search for the exact word, but look for words, sentences that are similar. So, for example, an analogy is an e-discovery search for defective product that also finds emails discussing a faulty device. So, embeddings make your AI smarter, if I may put it that way. Next.
Now, let's talk about ethics and duty. The ABA Formal Opinion 512 is the best closest tool or standard right now in the US and it's it's a national paradigm. So, it confirms that existing model rules are sufficient, but heightened diligence is required when using AI. 1.1, competence. Know your tools, know the AI tools, know its weaknesses, know its strengths, know its limits and the risk associated with it. 1.6 is confidentiality. First, I would like to say that avoid using free consumer AI tools because this means that you're putting your client's data or information at risk. And also when you're using consumer AI tools that you're paying for, ensure that they are not training their data. Rather, they're not training their model with your data, with your inputs. You have to ensure that and if need be, you could have contractual safeguards to ensure that your client's data and your data is safe. Confidentiality.
And next is communication. You have to, especially when using AI tools, explain and communicate this to your client and let them understand the related risks, the benefits and the costs. And usually, the costs of AI tools should ideally be an overhead of the firm's operation and should not necessarily be charged to the client. But of course, there exceptions to this on a case-by-case basis. Yes, so fees, reasonable fees. Be clear about AI-related expenses. I've almost touched a bit about fees previously, but I want to say one more thing. Based on the ABA Formal Opinion 502, 1.5, it is expected that you charge your client based on the time actually spent using the AI tool, prompting it, reviewing it, and not the time they would have actually been if you were to do the work manually. No. So, you actually charge for the actual time used with the AI tool, and not the time you saved, or not the time that you would have used if you were to do the work manually. And of course, this brings or ties to the topic of an or alternative ways of, you know, billing since uh from hourly to maybe value-based. Yeah, that's another topic. But, that's not the focus of this, so let me go on.
Supervision. 5.1 and 5.3. You treat AI like a non-lawyer assistant because it doesn't actually understand what it's doing. It's just basically predicting the next word based on statistical patterns. So, review and take responsibilities responsibility for the output of the AI because the court would not hold the AI software responsible, but would hold you responsible. Cander. This basically refers to AI citations, but not AI citations, but citations generally from the AI. So, you have to verify that the cases, the statutes from the AI is correct. You have to independently verify that by yourself. Yeah.
And now, let's go to what the state bars are saying. So, state bars reinforce the ABA principles, and the dominant framework is clear, which is treat AI as a powerful but fallible non-lawyer assistant requiring rigorous supervision under Model Rule 5.3 that we discussed earlier, so. 5.3. Okay.
So, let's talk about Florida. What's Florida saying? Florida emphasizes confidentiality, you know, it says informed consent when disclosing information to third-party tools. You need to tell your clients and ask for permission before disclosing their information to third-party AI tools, and rigorous supervision. Supervise the output of the AI tools because it is a non-lawyer assistant.
What is California saying? So, California's practical guidance says that the professional judgment of a lawyer should not be delegable. It's not delegable. You should always review the output of your AI. You must review, validate, and correct all AI output.
New York City provides guardrails and not restrictions. Says confidentiality, ensuring that your client's data are not used to train the AI model. Ensure that through contractual guarantees, and also conflicts. So, conflict here means that when, you know, you're having to put your multiple clients' data in an AI tool, um this might lead to conflict, especially if those clients might be adverse clients, competitors, or something like that. So, you have to also consider that and you know, look into that and see ways to, you know, mitigate that. And also, intake bots. This is kind of going into the realm of unauthorized practice of law, where you're using an AI tool to take in new clients without supervision, without human supervision. So, they also emphasize that you have to ensure that there's human supervision when you're using the AI tool to get in or intake new clients. And of course, cander. Ensure that you verify you verify the citations, the cases, the statutes that AI gives to you.
And in this slide, I will give you some AI use cases, and we're going to compare side by side the opportunity versus the oversight. First is document drafting and analysis. AI is quick to draft motions. It's quick to draft contract. But, the draft is only the first point. It's only the starting point. You as the human lawyer, the professional, you have to review it. You have to tailor it to your clients, and you have to verify compliance with the law.
Next is legal research. AI, of course, can go through large amounts of data in seconds, but you as a human lawyer, you have to verify every citation, ensuring that it did not hallucinate and conjure some false case laws or statutes.
Another is e-discovery and document review. Automated classification. As we spoke about, it can go through emails looking for not just a strict keyword of product defect, but looking at maybe a defective phone and all. But, you have to actually for e-discovery, you have to define the protocol with which or through which it does all of this, and ensuring that the protocol is valid, accurate, and of course, leads to the right and desired output that you want.
Translation, so. AI invariably can translate loads of languages now. And it's instant and low cost. But, for translations of documents that are really crucial important for cases, a professional has to review this and certify it.
So, here I am highlighting some tools that lawyers should know. Of course, this is not extensive. There's loads of tech legal tech tools out there. So, for research and drafting, Counseled, which is now a part of Thomson Reuters. Harvey AI. ChatGPT Enterprise. For document review, we're talking about Relativity AI. Disco AI. For contracting and practice management, we have Ironclad AI. Clio. And Spellbook to mention a few.
Selecting an AI vendor is not only an IT decision. It is a competence decision because it can have an adverse effect on your clients and even on your credibility if chosen wrong. So, here are some, which I wrote, a 10-point checklists for choosing an AI vendor based on the four National Institute of Standards and Technology AI Rules Management Framework. The first is internal policy. You ask the your AI or prospective AI vendor, do you have a documented AI policy? The second is accountability. Ask them, who in your organization is responsible for AI oversight and risk management? The third. Use case definition. What are the approved use cases for your AI systems? Fourth question is data sensitivity. Ask them, how do you classify and handle sensitive data? The next is accuracy mitigation. What processes do you use to test for and reduce inaccuracies, such as hallucinations? The sixth is bias mitigation. What methods or audits do you perform to identify and mitigate bias? The seventh. Data usage guarantee. You ask them, can you guarantee contractually that our data will not be reused for training? And the eighth question to ask your AI prospective AI vendor is on security certifications. Do you hold current security certifications? The ninth is on data deletion rights. Can you provide assurance and a process for secure data deletion? And the 10th is explainability. You ask your prospective AI vendor, "How do you ensure AI outputs are transparent, verifiable, and explainable?" This are 10 questions based on the NIST AI rules management frameworks that I believe every law firm or lawyer should ask a prospective AI vendor. And if they cannot answer any of these well or to your satisfaction, then you might hold off on using them because this are crucial. Remember, it's your clients that are involved here and your credibility. And you wouldn't want to jeopardize any of that. All right, let's go to the next one.
Now to intellectual property and policy watch. Let's talk about what the United States Copyright Office's guidance is. No copyright for purely AI-generated outputs. That's their clear stance. But where a human edits it, compiles it, or meaningfully shapes the outputs or outcomes of an AI-generated outputs, then that part can qualify for copyright protection. I'll say that again. Where a human edits, combines, meaningfully shapes the outputs of an AI-generated work, that part qualifies for copyright protection. But for a fully 100% or unadulterated AI-generated outputs without any meaningful human contribution, that does not qualify for copyright protection. And then there's a duty to disclose or limit claims for AI-generated materials.
Next, the United States Patent and Trademark Office inventorship guidance. An inventor must be a natural person. Clear stance. AI-assisted inventions can be patentable if a human had a meaningful contribution to its conception. Yes, I've said it. It requires a significant human contribution for an AI-assisted inventions to be patentable. Okay, let's go to the next one.
So, the EU AI Act timeline is the first major global standard for AI and it has extraterritorial effect, meaning that for US clients whose products are in the EU market, they also will be covered by the EU AI Act. And entered into force 1st of August, 2024. And February 2nd of 2025, the ban on unacceptable risk AI took effect. And in August 2nd of 2025, the rules for general purpose AI models became applicable with transitions for existing models. And next year, August 2nd of 2026, most high-risk requirements apply. So, for US law firms, lawyers, and US clients, it's time to begin to prepare, especially for lawyers, on advising your clients about the EU AI Act and its effect on your clients.
So, this is a bonus uh is a bonus slide. Um I just wanted to give a 30-60-90 day action plan. Uh if you've not started to use AI in your firm and you just want to start, yes, so I've given you I'm going to give you baby steps to start with to grow your confidence and eventually scale. So, the first 0 to 30 days for piloting. Next, 31 to 60 days, policy. And then 61 to 90 days is for scaling. So, let's go.
So, first, pick one workflow to test. For example, it could be drafting a research memo. Then define clear guardrails prohibiting clients' data in public tools. Now, based on what we've spoken thus far, you should have an idea on how to do this. Adopt a verification checklist for all AI outputs.
And the next for the next 31 to 60 days, draft an internal AI policy covering competence and confidentiality. Run vendor due diligence using our spoken NIST-aligned checklist. Then train the team your team on best practices for prompt retrieval and review.
And from day 61 to day 90 is for scaling. First is expand to a second workflow. Since you've gotten a good grasp of the first workflow, which is for example, drafting a research memo, so you choose a second workflow, which could be an e-discovery triage. Integrate a rag system over approved internal knowledge sources. If you remember what I spoke about when I explained rag, and then track quality and your time saved. Report value to partners and clients. And see where it goes from there. You might just love this AI if you're feeling adverse to it currently. Give it a try. Start with the safest workflow. Just within your system or within your firm. Doesn't have to go out. Doesn't have to be used in court. So, just try.
So, key takeaways. Competence is a non-negotiable. It's important to know the strengths, the weaknesses, the limits, and of course, the risks associated with this new technology called AI. Human supervision is mandatory. Treat AI as a fallible assistant, as a non-lawyer assistant, ensuring that every of its outputs gets human review. Confidentiality is paramount. Do not expose your clients' information tools that may disclose or use it, particularly free consumer AI tools. And then when using trusted AI tools, ensure that consent is obtained from your clients when necessary for exposing their before exposing their data or information to those AI tools. Vendor due diligence is a professional duty. It's not an IT decision only. Vet every AI vendor with a structured, documented process. AI is not here to replace lawyers. Instead, it's here to augment lawyers. But the only pathway is through safe, supervised, and ethical AI. Use AI to augment your expertise, but not to replace your professional judgments.
Okay, so, any questions? If you have questions because this now is going to be placed on YouTube and of course, on Spotify. Let me know your questions either in the comment section or send me an email to aiandlawbridge@gmail.com or you can connect with me on LinkedIn with the name Tolulope Awuyemi. And I will be glad to answer questions. Okay, thank you so much everybody and have a good rest of your week.