KENX AI in GxP East 2026

2026-09-16-0933-ai-misuse-regulatory-actions-lecture

30 min read

AI misuse risks and regulatory actions lecture

Recorded 2026-09-16 at 0933 via Plaud. 26:12.

Summary

A cross-industry tour of AI misuse and the regulatory reactions it provoked: biased credit screening, California SB 1120 barring AI as the sole decision-maker for healthcare claim denials, Department of Justice False Claims Act action against AI-driven medical billing fraud, biased applicant-tracking systems, deepfake legislation, and automated government systems that wrongly terminated benefits for 20,000 to 40,000 people. The recurring root cause is absent human oversight, which the speaker expects to remain essential for the next five to seven years.

He retells the Purolea Cosmetics warning letter (April 2, 2026) as the life-sciences cautionary tale, positions Computer Software Assurance as the risk-focused path forward (with digitalization first for the roughly half of the industry still on paper), and surveys the standards landscape: the EU AI Act as the closest thing to comprehensive regulation, Annex 22 delayed by at least a year, and ISO 27001 now a common vendor requirement. He closed on the FDA’s own use of Elsa for submission review: you cannot bring a knife to a gunfight.

Plaud AI notes

<App-generated notes from the Plaud cloud, verbatim.>

Date Time: 2026-09-16 09:33:37 Location: [Insert Location]: [Insert Location] Instructor: [Insert Name]: [Insert Name]

Plaud summary

The lecturer discusses the significant risks and regulatory actions associated with the misuse of Artificial Intelligence across industries such as finance, healthcare, and human resources. Key examples include AI models exhibiting racial bias in credit applications, automated denial of healthcare claims, and discriminatory hiring practices. The speaker emphasizes that these issues arise from biased modeling, insufficient human oversight, and rushing technology deployment without proper validation. The presentation underscores the need for accountability, transparency, and a “human-in-the-loop” approach, reinforced by President Biden’s executive order on AI emphasizing safety and human rights. A specific FDA warning letter to Purolea Cosmetics is cited as a case study of complete overreliance on AI and failure to comply with CGMP regulations. The lecture also covers the evolving regulatory landscape, including the EU AI Act and various ISO standards, concluding that while the environment remains a “wild west,” companies must adopt tools and processes aligned with those used by regulators to ensure compliance and remain competitive.

Knowledge Points

1. AI Misuse and Regulatory Violations

  • Bias in AI Models
    • AI models used for screening credit applicants have been found to deny minorities, indicating biased modeling.
    • A study is mentioned where a continuously learning AI became semi-racist, adopting derogatory terms learned from data originating from specific regions of the U.S.
    • This necessitates continuous validation and monitoring of models to detect drift and bias that lead to improper results.
  • Misuse in Healthcare and Insurance
    • In California, legislation (SB 1120) was passed because insurers used AI to automatically deny healthcare claims without human oversight. The law now requires that AI cannot be the sole decision-maker for coverage denials.
    • New regulations mandate that healthcare providers must clearly disclose when generative AI is used in patient communications.
    • The Department of Justice is targeting AI tools used in medical billing that submit false data to federal health programs under the False Claims Act, analogous to other insurance fraud.
  • Bias in Employment and HR
    • Applicant Tracking Systems (ATS) have been used for decades, but adding an AI layer can introduce significant bias, such as discrimination based on gender or race.
    • The speaker stresses that inefficient or biased HR data sets waste time and resources across the organization.
    • New laws, such as one in North Dakota, are being introduced to protect labor, forbidding AI systems from violating collective bargaining agreements or causing unlawful employee displacement.
  • Social Media and Data Privacy Violations
    • Major LLM companies have had to adjust their systems due to sexually explicit deepfakes using celebrity photos, raising issues of privacy and protection of likeness.
    • California law (SB 981) now requires platforms to remove such deepfakes.
    • Companies have been penalized for failing to secure user data during AI training or for using personal data without consent.
  • Failures in the Public Sector
    • Automated government systems wrongly accused 20,000–40,000 citizens of fraud, leading to wrongful termination of benefits.
    • Faulty facial recognition technology used by law enforcement resulted in wrongful arrests.
    • School districts using AI monitoring tools faced scrutiny after accidental leaks of private student data.

2. The Need for Human Oversight and Proper Process

  • Human in the Loop as a Core Principle
    • A consistent theme across AI failures is lack of human oversight, accountability, and accuracy. The goal must be building confidence in outputs.
    • The speaker predicts that for the next 5–7 years, human oversight will remain critical. Autonomous AI will be adopted only when modeling, validation, continuous monitoring, and change control processes inspire high confidence.
    • Current issues stem less from AI technology itself and more from the need for human oversight and adherence to existing regulations.
  • FDA Warning Letter to Purolea Cosmetics
    • A warning letter dated April 2, 2026, was issued to Purolea Cosmetics for inappropriate use of AI in pharmaceutical manufacturing.
    • The company used AI to create drug product specifications, procedures, and master production records but failed to have a human review them for accuracy and CGMP compliance, violating 21 CFR 211.22(c).
    • The company failed to conduct process validation, claiming the AI agent never indicated it was required—demonstrating dangerous overreliance on AI and lack of basic regulatory knowledge.
    • The FDA mandates that if production resumes, any AI-generated output must be reviewed and cleared by an authorized human representative from the Quality Unit.
  • Computer Software Assurance (CSA)
    • CSA is suggested as a path forward. It is not a shortcut, but a way to focus validation on high-risk areas.
    • It enables companies using cloud technology to leverage vendors, improving efficiency by prioritizing critical areas, bug fixes, and historical failure points.
    • Digitalization is a critical first step for the roughly 50% of companies still on paper, preceding automation and AI.

3. Evolving Regulatory and Standards Landscape

  • Current State of AI Regulation
    • The speaker notes there is currently zero official AI regulation in the U.S., describing the situation as a “bit wild west.”
    • The EU AI Act is the closest to a comprehensive global regulation and is highly recommended reading, with focus areas relevant to life sciences.
    • President Biden’s executive order on AI emphasizes human rights, safety, security, and traceability across government branches.
  • Key Regulatory Guidances and Standards
    • FDA AI Draft Guidances: Highly recommended, especially for medical devices, covering monitoring, validation approaches, and encouraging early FDA engagement.
    • Annex 11: Focuses on cloud technology, cybersecurity, and vendor network evaluation, aligning with CSA.
    • Annex 22: Expected to be delayed by at least a year due to extensive industry feedback.
    • ISO Standards: ISO 27001 (data security/privacy) is now a common vendor requirement. Other standards like ISO 13485 for medical devices provide frameworks, while AI-focused standards (e.g., ISO 42K series) remain nascent without a single clear pathway.
  • The “AI Arms Race”
    • The FDA is using AI (e.g., “Elsa”) to review regulatory submissions, accelerating reviews and identifying gaps.
    • The analogy “you can’t bring a knife to a gunfight” underscores that companies must adopt similar AI tools to pre-evaluate their submissions.
    • AI can assist human auditors by finding gaps they might miss, enhancing accuracy as well as speed.

Questions

  • [Insert Question/Confusion]

Assignments

  • 1. Read the EU AI Act to understand the most comprehensive existing framework for AI regulation.
  • 2. Review the FDA’s AI draft guidances, particularly those in the medical device space, to understand regulatory thinking on validation, monitoring, and security.
  • 3. Evaluate internal processes to ensure adequate human oversight (“human in the loop”) for any AI systems currently in use or under consideration.
  • 4. For companies still using paper-based systems, create a digitalization plan as a foundational step toward automation and AI adoption.

Highlights (button-flagged moments)

Regulatory Actions Targeting AI Misuse in Healthcare and Insurance

While displaying a slide detailing AI violations in healthcare, the speaker discusses specific regulatory responses, such as California’s legislation against automated claim denials and the Department of Justice’s focus on AI-driven false claims. • Visual Context: The presentation slide outlines three key regulatory actions in healthcare/insurance regarding AI: restricting automated claim denials, mandating disclosure of AI use in communications, and targeting AI-driven false claims. • Discussion Insight: The speaker highlights a common theme across these examples: the need for accountability, accuracy, and human oversight when AI is used for critical decisions like insurance claims. • Key Takeaway: The use of AI by ‘bad actors’ to automate decisions without proper oversight or to commit fraud has directly led to new legislation and targeted enforcement to ensure accountability.

Discussion on AI Violations and the Call for Human Oversight

Using a slide that outlines key AI violations in hiring, labor, media, and data privacy, the speaker elaborates on these real-world examples to emphasize the widespread issue of improper AI use and the urgent need for human-in-the-loop accountability. • Visual Context: The slide presents four key areas of AI regulatory concern: bias in hiring, labor protection violations, deepfakes, and data privacy breaches. • Discussion Insight: The speaker highlights that many of these violations, such as biased algorithms, are not necessarily intentional but result from a lack of human oversight and continuous monitoring. • Key Takeaway: The central theme is that until AI models achieve proven confidence and robust validation, human oversight is a critical and non-negotiable safeguard, a sentiment reflected in emerging regulations like the EU AI Act.

Speaker Elaborates on Slide’s Examples of AI Misuse and Negative Societal Impact

The speaker directly discusses the on-screen examples of AI failures, including wrongful fraud accusations and faulty facial recognition, to illustrate the real-world consequences of biased and unregulated AI systems. • Visual Context: The slide lists three key examples of AI violations: automated systems wrongly accusing citizens of benefits fraud, faulty facial recognition leading to wrongful arrests, and AI monitoring tools leaking private student data. • Discussion Insight: The speaker explicitly references the fraud example, noting how automated systems wrongly terminated benefits for 20,000-40,000 people, highlighting the severe impact of AI errors even within government. • Key Takeaway: The discussion uses these specific, visualized examples to argue that across all industries, a lack of human oversight and regulation for AI is leading to significant negative outcomes, such as privacy breaches and biased decisions, making human-in-the-loop systems essential for the foreseeable future.

FDA Warning Letter to Purolea Cosmetics Highlights Risks of AI Overreliance and Lack of Human Oversight

The speaker uses an FDA warning letter to Purolea Cosmetics, which details the company’s failure to validate processes because they blindly trusted an AI agent, to exemplify the critical need for human oversight and accountability in regulated industries. • Visual Context: An FDA warning letter dated April 2, 2026, cites Purolea Cosmetics for CGMP violations, including failing to conduct process validation and not reviewing AI-generated compliance documents. • Discussion Insight: The speaker frames this case as a quintessential example of a company prioritizing speed over regulatory compliance, stating they ‘just trust AI one hundred percent’ and ‘don’t even know the regulations’. • Key Takeaway: The FDA’s conditional requirement for Purolea Cosmetics to implement human review if production resumes reinforces the speaker’s main point that for the foreseeable future, human oversight is essential when using AI.

Speaker Discusses FDA’s Computer Software Assurance (CSA) as a Framework for Modernization

While displaying a slide of the FDA’s guidance on Computer Software Assurance (CSA), the speaker explains that adopting CSA is not a shortcut but an efficient method for companies to modernize processes by focusing on risk, which is a crucial step toward digitalization and eventual AI implementation. • Visual Context: The screen shows the cover of the FDA’s ‘Computer Software Assurance for Production and Quality System Software’ guidance document, which applies to systems used in medical device production. • Discussion Insight: The speaker clarifies that CSA enables companies to advance processes efficiently by leveraging technology vendors and focusing validation efforts on high-risk areas, rather than skipping steps. • Key Takeaway: Adopting principles from frameworks like the CSA guidance is presented as a foundational step for improving quality culture and digitalizing processes, which is necessary before an organization can confidently move towards using AI.

Transcript

[00:00] Speaker 1: That’s giving the right results. Maybe you want to be old school for the moment. So, if we look at some other areas, and I thought these were interesting, especially because I spoke to you about yesterday about being a part of the the executive order and President Biden on AI. That was that was a lot of this, but that was theoretical, theoretically, because we really hadn’t came yet. It was more, Providing the initial foundation. If you look at this, talks about denying credit to minority applicants. They used AI to screen and automatically, maybe the the modeling was biased, right? I think there was a study several years ago. One of the big companies developed an AI, and they were they were continuously learning, and suddenly it seemed that some of the data was. Happen to be from certain states, certain regions of the country, and the AI was starting to utilize, become almost semi-racist. It was using derogatory terms as what it thought was acceptable based on some of the the volume of responses from certain parts of the U.S. Right. So similarly, the AI in this case is already now denying applicants based on based on race. Right. So. That means that means it could be any example that I put up on the screen. It means that model has to be looked at. It has to be. It may have been fine in the beginning, and suddenly the modeling of training has keeps occurring. It means that validation, the drift, the bias, where it starts giving improper results, has to be filtered. And what Dory mentioned yesterday too. If you look at the CEOs of large LM companies, that’s what they’re all saying now because of. Because AI has escaped itself, it’s attacking companies. Whether that’s true or not, we don’t know. I’m sure half of that is only true. I’m I’m sure that people steal people’s stuff all the time, and they just don’t tell us. Right? Governments do too. And but that’s the reality of life. And so they’re also looking at you know more more corporate activities. Conversations where they want to stifle stifle competitiveness, right? Because they want to get the market share at the vote, right? You know that happens every day for for decades. Similarly, if you see here using automated denials for healthcare claims, we found in California with some legislation passed because they’re. Because they were utilizing AI to basically make a decision, so different industry, different use case, but you see a similar theme going on, right? It has to do with accountability, accuracy, and and what we’ve done for a long time. Do you feel confident in the output? Like in this case, insurance company may love it because it’s quicker, faster. Probably AI is saving them money. But the reality is that that it may be misleading, that it’s not performing properly. Maybe it’s and and there is no human oversight, right? So in this case, this is the example. If you see, it talks about that it being clearly disclosed for healthcare providers that they’re using AI in their process, right? There’s a False Claims Act as well, where the Department of Justice targeted AI tools in medical billing that were submitting false data. I mean, insurance scams, fraud, collecting money. We have NBA, past NBA players that create similar companies, and in the last year, that are doing the same thing. Not with AI, but but now they’re utilizing tools for bad practice, right? And so. Unfortunately, in life, regulations and and acts and laws are put into place. Why? Because there are bad actors. The reality of life is that there is there’s lots of companies, whether in farm or another industry, that that falsify, delete data, fraud, even today. Right? That supply. Probably, I mean, sometimes I look at my medicine cabinet, and I, you know, it’s. The The scariness of it is that when you see too much over time, sometimes you don’t trust anything, right? And but you have a choice, but you have limited choice to some. In the next example, it talks about human bias about hiring practices. And how many how many people in the room have a automated applicant tracking system and Even a filtration process that gives you the top five percent of your candidates, but we have an applicant tracking system that’s somewhat automated. No, but he’s in HR. You still, I know. You might know your boss, right? You may be the person that interviews. So, so many companies, especially big ones, have been using it for decades already, right? Now, now imagine like adding a layer of AI to it, and suddenly. Yeah, has a bias to it, and maybe because it’s HR, that’s it’s not the GXP or the the thinking is is different, right? And so that becomes important because how many people? Let me ask another question: How many people here are responsible for doing technicalities in the organization? Okay, a lot of them. So, so that initial data set is critical because. Then you’re getting either garbage, or you’re getting something you didn’t ask for, or something totally wrong, right? Then it’s a waste of time. It’s a waste of your time. It’s a waste of the organization’s time. The inefficiencies of a process have to be looked at, not the first step, the whole cycle, the whole cycle, because it may look pretty in the front, and then you have damage in the back, right? So, so ultimately, as you can you can kind of look through some of these examples, but. You know, unlawful employee displacement. You look at AI-generated images. So some of the major LLMs have tweaked it because people are taking celebrity photos, and that’s a whole other separate issue. And and putting different kinds of gloves on people, and and there’s there’s privacy, there’s protection of people’s likeness. People are some of the people that are movie stars, and they have various protections. Let’s say. But ultimately, what you’re seeing is also a huge kind of gap in data and looking at data privacy violations. So, so if you look at the summation of of what’s out there today in terms of government, there’s a huge focus. the the The executive order, President Biden, also, if you remember what I said yesterday, it focuses on on human rights, focused on safety, and and it highlighted. It highlighted this, and that was across every single branch of the government. Defense, they don’t share that stuff. It’s private, so I couldn’t tell you what that was. But, but more or less, there because you know, information is. We are an information society. That means that data has a value. The personal data has a value, and people use it if they have access to it. Sometimes. You can imagine using AI as you saw in some examples for clinical, right? There’s a lot of data there. There’s a lot of money that can be had if I steal that data. So, if you see here, this example talks about fraud. Automated systems have wrongly accused citizens of fraud. Twenty to forty thousand cases in one instance were caused wrongful termination of benefits. So even the government has to be careful of it. And the American Society for Quality has a government division, and they look at you know what the policies are, the use of technology as well. And there’s a lot of cross-functional seeding of proliferation between groups from the software division to the government division to the regulatory division to innovation to to things like this, right? That are being utilized. Facial recognition. Seen enough movies, doing it for a long time, probably will tell you what they do, right? I think during COVID in China, I heard police officers had visual face masks, like face masks, like a Terminator movie, where they can look at you and see what your status is, if you’re a criminal or not, whether you’re supposed to be allowed on the street when there was lockdown policies, and so it’s it’s like Arnold Schwarzenegger looking at you, and he can quickly assess you, right? So. So, but there has to be some protection here too, because maybe you get, maybe you get your treatment as a human being may become biased based on what data is shown to a police officer in another country. So, even if you’re visiting, universities had this example of university student data getting privately leaked out. I mean, it didn’t disclose what that data was and what the intentions behind. May have been, how people are misusing it, but but so I think what you’re seeing is you’re seeing across cross scrutiny of of areas. We’re we’re seeing lack of transparency overall. We’re seeing a lot of data privacy breaches, and ultimately they’re all important, but the bottom one really shows you that there’s bias in the algorithm, right? I don’t. In some many of these cases, I don’t think they’re purposely trying to do that. But nobody’s paying attention. There’s there’s no human in the loop. If that if if we were sitting here today at the American Society World Conference that was in Orlando earlier this year, I’ll tell you I was presenting different regulatory observations in in different industries outside of farm as well. And the bottom line is that is that there’s there’s a lot of. Improper AI that’s out there today across all the industries, and so so this becomes the fear that it’s being used incorrectly. It may be efficient for you, but the result, the output, is is and it has negative effects too. So you see here, right to human interaction, human in the loop. Right, and this is about public services. Right, so that’s the that’s the trend I see. Is that for today, for the foreseeable future, the the human oversight becomes important. One day, there will be confidence in the modeling and the validation and the testing and the output, and our documentation, and the continuous monitoring of those processes, and the change control within it throughout its process. Then, at some point in the future. Can be five years, can be seven years. I can’t predict. AI couldn’t tell me either this morning. So I think in the future, depending on organizations, there will be some too. There will be some autonomous AI that makes a decision. But it’s a matter of risk. It’s a matter of confidence in the process. And of course, today we have zero regulation. The only thing we have is the EU AI Act and some things that already have taken place and are enforcing. It’s coming. This really, to me, is not about AI. It’s about it’s about human oversight. It’s about following regulation. Many people have discussed it. Many, I’m sure, everybody. How many people have read this morning? Let me raise your hand. Less than I thought. It’s it’s interesting to see the evolution. To again, my opinion only. I’ve worked in different companies. I’ve been inside many different plants around the world. I have seen, I’ve seen the difference. Generally speaking, shouldn’t be, but I see a quality step down. If you’re walking down a staircase, I in this order doesn’t mean the the branded can’t be bad. There’s plenty of branded companies that have branded products that are innovative that go under consent decree. What it means is, generally speaking, I have seen if you go from branded to generic to OTC to consumer health, I don’t think the quality standards are the same. Not nearly even close to being the same. The technology is not the same. The culture of the quality culture of the people is not the same. It becomes in some of these cases down low. It becomes because you’re making a penny or a half a cent one one product versus. Last night and today, we’re talking about some East Coast companies that not doing so well because they don’t have the population of patients to support the commercial product. But the the initial approval was three billion and four billion dollars for two of two FDA approved products for one dose. So there’s very few people that can actually pay for it, and so so the issues is that. This example typifies doing something fast and not caring about anything else. That’s it. Plain simple. You don’t follow regulation. They were they were stupid. They they don’t even know the regulations. They just trust AI one hundred percent. They just they want to do it fast and get a result, right? And they have pharmaceutical, they have cosmetics, they have other types of products. So probably. The management of that company operates in a very low-cost margin type thing. Most of the regulations and most of are either similar or the same. It doesn’t matter. Imagine the CDMO or contract testing lab that you employ doesn’t qualify or validate any of your systems. I can tell you in the last three years, I’ve seen companies, and I, I, it still baffles me. How they ever pass even any customer audit, and you know they have many customers, so they have many audits. So then it really, it really makes you wonder the quality of the auditors coming in, because twenty companies. It’s how can you just right in in hours without even doing an audit, just by training people and and kind of asking some questions. You can kind of tell without even looking at. Any form of paper. So I only put CSA up there. It’s my own opinion. I think for some companies, again, remember, 50% of the world is still on paper. That means that you have to digitalize. You have to walk up the staircase. You have to digitalize. Then you have to come up with automation and using tools, develop the culture. CSA allows you if you’re using cloud technology. You want to leverage your technology vendors. You want to you want to make your process efficient. Don’t waste your time to do things that you’ve done for ten, twenty years when you can advance that quickly. Doesn’t mean it’s a shortcut. It means it’s it’s or it’s skipping things. It means that is you spend your time where you think the risk is. You have a history of a performance. Most of you are using products you’ve used for a long time from the same manufacturer, maybe a different version of it. And so, utilize that data. Where you have issues, spend time there. Where you have failures, spend time there. When you’re looking at a bug, bug fix release, spend your time there. Check what’s critical. Depending on what the overall system’s output is and decisions, decisions that are made based on that data, then you can have steps to AI. Right? It doesn’t mean there’s not band-aids. There are plenty of companies that show you can scan a piece of paper. They can start analyzing. That’s a bad thing, right? We can do that, but there’s also faults in the scan. Someone misses a page, it only copies half the page, whatever it is, right? And I wouldn’t want to be the guy that has to scan two hundred pages. Okay, so overall, if we look at some of the summary, that the future, the future is digital. The future means that analytics have to be informed. Whether they update Part Eleven or not, it’s looking at a least burdensome approach, but it’s refocusing our interests. Ultimately, if you see the various draft guidances, they’re very good. Honestly, if you read some of the, if you have not read some of the FDA AI draft guidances in the medical device space, I highly recommend it because they’re looking at they’re looking at kind of they talk about monitoring, they talk about a different validation approach. They talk about, and some of them in manufacturing that have been released talk about. I’m sorry, drug development talks about having those conversations with the FDA about it. It doesn’t mean you can’t use concepts from others, even if you’re not making a medical device. Use. I encourage you to use some of those concepts, right? Because it’s still the thinking of the government, right? It may be for a different niche within our industry, but ultimately. As you’ve heard today, it’s it’s talking about the security of the framework, and and even in Europe, already, you know, we’ve taken a turn out a lot of it, right? Some companies in the U.S. and read this off-the-wall materials because I can tell you, the if you have an AI tool, it can be ten pieces, ten pieces that you put together, and then behind the scenes, it could be twenty different companies that have put that product together, right? And that becomes important because they want. Traceability and transparency, as well as evaluation of that network of vendors, right? So, Annex 11 is basically focusing on cloud technology, similar to CSA, similar to ISO 13455 updates, and also cybersecurity, right? Annex 22, I would say, is probably going to be delayed for till next year sometime. It was supposed to probably get done this year, but. With the industry and regulatory meetings and the amount of feedback, it’s at least a year delayed. So the EIA Act is quite good. It focuses on different life science components of the industry. I highly encourage you to read it if you’re not. But it’s the closest we have to something official in the world. That has good. I think we’re right about time. I wanted to leave a couple minutes open in case anybody has any questions that you’d like to ask or ask me.

[20:30] Speaker 2: There’s been a nice suite of ISO standards about that. We have management system, risk management, things like that. In the same way, for medical devices. 1345, which is fact, USR they confirmed, and so you were kind of always safe if you did 1345. Is there any safety like that with the standards that exist now as they can converge with the.

[21:01] Speaker 1: Regulations? I mean, one most companies like I know we have accreditation. A lot of organizations last two years have required their vendors to have ISO 27001. Right, it’s really about the security and the privacy of the data and the connectivity. Right, oftentimes clients are like, “Oh, I only want you to use my system,” which I kind of shake my head because if you’re accredited and you’ve been checked, but you have a certificate, standards certification, that what’s worth the distrust? I get it. I mean, there’s always vulnerabilities and the landscape changes. There’s several ISO standards that are out there, depending what what niche you’re working in, what industry that that go to. AI.

[21:44] Speaker 2: Specifically for AI, I like the 42K, which is advanced stuff like that. It’s kind of the wild west right now.

[21:53] Speaker 1: Or is there like some stability and safety? Yeah, I would say it’s a little bit wild wild west because you have new standards to do in AI, so there’s a lot. I think what people people don’t know is is what do I follow, right? And so some people cite this number, some people cite that number, some people have nothing; they create their own because, There’s no regulation, at least in this country, so it’s a bit wild west. I mean, I mean, you can look. Remember, you can follow whatever you. I mean, it’s in today’s world. It’s up to you, right? What you want to follow, but and it depends on how you write what you follow. If you say I’m following CSA and FDA, CSA draft guidance, then you can you can be inspected, even though it’s not a regulation. It’s not the inspectional standard. It’s not so. So your ISO standards depends on what industry you have, because again, it’s a standard. I mean, in pharma, in pharma, you may have nine thousand one, you may have two thousand, you know, two seven zero zero, and you might have thirteen eighty four, four eighty five. But other than that, you know, we’re mostly fine-tuned to regulations. So they they help to assist in thinking of of specific niches areas and how to create some framework. So. If you’re producing software, it’s a different scenario as well. They may follow some of those standards to show quality of the process. Any other questions? Thank you very much.

[23:28] Speaker 3: We all know that FDA is using Elsa for the regulatory inspections, but can you please elaborate based on your experience? How exactly are they using it? For example, will Elsa be integrating with Ready Room when we upload some documents as part of the request?

[23:48] Speaker 1: I honestly can’t tell you because I don’t work for the FDA and I don’t use us. So I mean, I mean, they’re using it for review of regulatory submissions, right? So speed up the review process to identify gaps in the submission. It’s no different than the industry side of things. Some companies have an AI regulatory AI that does the same thing before you submit it. What I would say, very simply, is this: War is not a good choice of words, but this is a race, and the race also means that you have to play. If you are on the battlefield, what do they say? You can’t bring a knife to a gunfight, right? And so you have to think about what you’re doing, what industry is developing, what what your competitors are doing, and what the government’s doing, because in the end. If you if you bring a horse to a a Lamborghini race, you’re not going to win that race. So what does it mean? It means that you need to have similar systems that are evaluating and helping the human before you submit that document, because they’re already going to utilize it to see where the deficiencies are. Similarly, as I mentioned a couple times, there are AI that are in different pilots within the government for medical device inspection. For for different areas where again, you know, maybe in the future it’s commercialized within the government where they’re utilizing the tool this as the human. I gave the I’ll just close with this. I was requested by a major company to bring an FDA FDA inspector with me on to do an ISO thirteen forty five. Not only the immature culture of one particular unit, but also the acquisitions that recently. Because they felt that they’re not up to par, they want to have a dual-headed monster that will evaluate and be able to deliver. The AI finds some things that the human doesn’t find. It’s not just about speed; it’s about the accuracy and about being able to find gaps that maybe I wouldn’t find. It may assist the auditor as well because not everybody has an expertise of everything.

Graph View

Backlinks