The AI Risks Leaders Need To Know About Today

This is the transcript of The Refreshing Leadership Podcast episode: The AI risks leaders need to know about today, published on 8th June 2026.

Right now, organisations all over the world are having the same conversation. How do we use AI fast enough to stay competitive without creating complete chaos? We have some leaders cautiously experimenting. We have other teams secretly using AI through VPNs because they feel they can't keep up without it, and yet they're not permitted by their organisations.

At the same time, we see growing concerns around data leakage, security, AI hallucinations, over-reliance on AI, and people outsourcing more and more of their thinking.

Despite all of this, my guest today is deeply optimistic about the future of AI. In part two of this conversation with AI governance and cybersecurity expert Dr. Adeel Shaikh Muhammad, we explore why he believes AI could massively improve cybersecurity and scientific progress, the hidden risks that leaders and teams are underestimating, why garbage in, garbage out matters more than ever, and why we should be thinking about AI more like a powerful intern - but not the final decision-maker. We also look at shadow AI inside organisations and why critical thinking is becoming more valuable today, not less.

Maya: Adeel, welcome back to the podcast.

Dr. Adeel Shaikh Muhammad: Thank you, Maya. Thank you for having me again.

Maya: When you think about everything you're seeing in this space - the possibilities but also the risks, especially on the cybersecurity and governance side - do you sit on the optimistic side, or the quite nervous side, or somewhere more neutral?

Dr. Adeel Shaikh Muhammad: I am optimistic in this AI space, because of amazingly great new tools which are not only saving time and increasing productivity and efficiency, but also keeping us safe. Just like Claude introduced Mythos - that is one of the greatest tools in cybersecurity history according to the analysts. It's helping a lot in terms of securing applications and identifying vulnerabilities. It's a powerful tool and they're not releasing it right now, but it will be amazing for us as cybersecurity consultants to secure organisational environments, if we are using it in a safe way. So I'm really, really optimistic about the future of AI.

Maya: I did hear about Claude Mythos. It's incredibly powerful, but they haven't released it yet because it could be dangerous in the wrong hands?

Dr. Adeel Shaikh Muhammad: Exactly. Because ethical hacking is still hacking, right? We as cybersecurity professionals are on the ethical hacking side, where we try to hack environments just to see how good their boundaries are - to test the waters, not to break things or do something wrong to an organisation. Whereas the other side are the real hackers whose intentions are wrong, the adversaries. So we need to do everything in an ethical way. That's why Claude Mythos could be a dangerous tool, but it's also super, super powerful.

Maya: I think it's great that you've mentioned the power of the security tools that are now coming forward - that's probably what's making you more optimistic. We have seen such a fallout in the last few years, some really high-profile hacking and cybersecurity issues. So if you're seeing that AI can really do incredibly well on that front, it's very positive. I feel similarly about climate change - that there are going to be things blown out of the water in terms of technological advance. Is that in your realm too?

Dr. Adeel Shaikh Muhammad: Yes, in every field of science - medical, climate change - the possibilities are endless. AI is doing advancement in any field of science you study because it's a great tool to think, a great tool to research, a great tool to build. It raises the bar of where we are right now as humanity in that particular field or niche. And the same thing happened when the calculator came along - people were saying, what will happen to maths? How will we study? So we increased the bar.

We introduced calculus, which the small calculators of that time couldn't solve. That's how we keep increasing the bar in every field of science. And that's good news for all of us - we can do more advancement, more research, and come up with solutions which never existed before, whether in climate change, cybersecurity, medicine, or other areas.

Maya: And alongside that, you're fighting the good fight by working in security and governance - which is where there can be risk and failure and gaps that need to be properly addressed. You're optimistic, but you're also doing the important work of addressing those risks.

Dr. Adeel Shaikh Muhammad: Thank you so very much. Yes - in every war you are seeing, there is a part of cybercrime and cyber war. We are in defence, and that defence is about securing all the digital boundaries instead of physical boundaries. So we are in a war zone, and we are trying to make this world a more secure, better place.

Adeel's personal use of AI

Maya: I'm curious on a personal level - some of the best things AI has done for you in the last couple of years, or how you personally use it?

Dr. Adeel Shaikh Muhammad: As we were discussing in part one, try to work out what tasks you do every day and try to automate them. I have built up to now 20-plus tools just for myself - small Chrome extensions, small applications just for my PC or MacBook - which are helping me out. It's not only about replacing manual summaries or repetitive writing. AI can give you so much beyond that. So build.

Maya: Can I probe you for some of those tools? Anything that might be a bit accessible for the rest of us or of interest?

Dr. Adeel Shaikh Muhammad: I have created a lot of skills on Claude which can help do multiple tasks, and I think those are available on my apps as well. I will share the link - they are available for free download and anyone can use them. They cover things like building images, marketing, or sales calls. Sometimes I need to do warm calling to someone I don't know much about, and AI is helping me a lot there. I have a small application that integrates with Apollo - which is a very good sales tool that pulls information from LinkedIn and other sources. I also use Deepify a lot to automate certain things.

And I've already created my digital twin using HeyGen, for social media content and short videos. I'm trying to build these tools using AI, even though I don't know much about coding - just to help me. And if everybody is trying to do this for their own teams, even within a bigger organisation, that can really accelerate things and have a great overall impact.

Maya: That's really interesting. I was hearing about a busy parent who had created a specific inbox that an AI agent could use and send emails from, related to school matters - to help consolidate things because she was receiving too many. So there are life hacks as well, aren't there. As somebody juggling work demands and family life, sporty kids with busy schedules - is there anything that springs to mind that I could try?

Dr. Adeel Shaikh Muhammad: This is a very interesting time for SaaS applications, because what we're discussing right now - anyone can build those tools. A little tech-savvy person can do it, and that's the future. Everybody is using five or six tools tailored to their own needs. But first, try to solve the problem for yourself. Once it's doing the job amazingly well, then you can try to sell it or help the community. OpenClaw also started the same way - someone was just trying to solve their own problems and build small use cases where AI could help. So going back to the same point: build use cases. Once you know what you need to solve, it will be much, much easier.

Maya: And on a personal level - as a coach I currently use ChatGPT, and I find it excellent across all different areas. I have a gardening chat, a health and nutrition project, all kinds of projects. In whichever field you're in, it can be incredibly powerful. I also think it's really powerful from a legal perspective - getting my trademark, it helped me a lot, and I think I would otherwise have needed legal support. I'm finding it to be a very all-round assistant, though I probably need to branch out into other tools now for specific things.

Dr. Adeel Shaikh Muhammad: I always prefer one tool over five tools, because now ChatGPT knows your preferences and your style. Sticking with that is a great choice if it's completing everything you need. And now after Codex and others, it's pretty much on a par with Claude - doing the same things. So if you can stick with one tool, I'd prefer that.

Maya: Yes, exactly. And just one more on tools - because I think this is one for the busy leader - you should be getting a summary of your industry news in the morning, right? A summary of what your day has to offer, or what would be a productive use of your time. My C-suite clients in large organisations aren't necessarily doing this yet. When it comes to organising and planning your days and weeks with AI, do you feel people can better use these tools to get ahead? Because if you outsource some of that thinking to a tool, there's a risk of losing some degree of control and thinking. What's your take?

Dr. Adeel Shaikh Muhammad: When we're talking about planning or just getting a summary of what's going on in the market, I don't think there is any harm in that. Because - and I'll echo this - humans should always be in the loop. We are not asking AI to make decisions. We are not asking AI to step in and do something. It's a small scheduler. For example, you can create a skill so that whenever you type /goodmorning, it runs automatically and gives you your summary. Or if you're using a tool that doesn't offer scheduling, you can make a PDF and copy-paste it in every morning, and it will give you whatever that PDF is asking for. I don't think there is any risk in this, unless you are asking the tool to take actions without your permission.

Maya: Like purchasing things, and so on.

Dr. Adeel Shaikh Muhammad: Exactly - purchasing, or controlling your PC. That's something I'm seeing a lot right now. Two or three years ago, when Microsoft tried to take screenshots of your browser and use that context to help with Copilot, everybody was scared. Now we are in 2026, where OpenAI can access your whole computer and go anywhere. And I have seen people just saying yes, yes, yes - give it full permission - which is really, really scary.

The risks leaders are underestimating

Maya: So now we're moving into the risks piece. Where do leaders need to be more cautious? What are the risks around data security that people are underestimating? If you were to give people a starting point - the minimum security measures they need to have in place, or three to five things to be thinking about - what would that be?

Dr. Adeel Shaikh Muhammad: Let me give you the example of API keys. An API key is everything. If you give someone the API key to your Gmail, for example, they can access your entire Gmail account.

And people are copy-pasting API keys everywhere - I have even seen them shared on social media. As you mentioned, we are not trained on this. Most people don't even know what an API is, or what MCP is. So please try to understand that. And number two - you can use the same chatbot to ask it for precautionary measures. Ask it: tell me how I can fine-tune my boundaries. It will help you. There are hundreds of YouTube videos on how to secure OpenClaw how to increase guardrails.

The top overall risks I can mention are: first, data leakage - sensitive information being pasted into these tools. Second, hallucinations - confident but incorrect outputs, and that's happening a lot. And third, over-reliance - where you are delegating your thinking as well. Like copy-pasting every email and asking ChatGPT to reply.

From the cybersecurity angle, we need to define what data is allowed and where validation is needed. AI is basically powerful but not accountable. I always give this example: in cybersecurity, if an AI has taken a decision, in the morning there will be a blame game. The SOC manager will shout, the cybersecurity manager will shout - who made the decision? And there is no one to take the blame. So trust AI like a smart intern, but not the decision-maker.

Maya: I love that. You've got somebody with all the degrees - like an MIT graduate - but they don't yet have the life experience, and therefore they're like an intern in many ways. That's a really good way of thinking about it.

So you've given us a lot of food for thought. If you are extensively using a tool, ask it how to protect and secure yourself. You've mentioned data leakage - and as you said, people can be really cautious initially, but once they've started using these things and they're in a rush, they start clicking through and giving permissions, and you do need to be really careful there. There are settings even in ChatGPT that you can turn off from the beginning, in terms of what of your information is being fed into the overall model. So there's a lot you can do in the setup of any particular tool to protect yourself.

Dr. Adeel Shaikh Muhammad: Exactly. Every tool has this option, and you can ask the same tool: how can I secure myself, how can I turn off permissions that aren't required - especially once you've made connections to your browser or Google Chrome. That's a big thing. And there are other issues like shadow AI - where your team is using different AI tools that aren't allowed in your organisation. The major business impact is data leakage, which can lead to IP leakage - which connects to what you were discussing about trademarks - and to client confidentiality breaches, which can cause regulatory exposure including under GDPR, NCA, SAMA, NIST frameworks, and ISO standards.

Maya: So based on what you're saying, it's worth investing time upfront on any tool to get yourself set up as confidently as you can - or taking time out once you've started using a tool intensively to go through that checklist. That's more on a personal level, and I know organisations are really hot on this and will have their own internal procedures about what's allowed. But that's really helpful.

Before we wrap up - at a team level, where there might be resistance or difficulty engaging with AI, any tips on how to bring everyone on board? It could be age-related, or people who see themselves as being in the final stages of their career and feel it's not for them.

Dr. Adeel Shaikh Muhammad: Training and awareness. If an organisation doesn't invest in that, they are missing out. Try to explain it - everyone needs to learn. So instead of letting people go, just train them on AI and ask them to do two, three, four times what they are currently doing. That will help them get on board - from prompt engineering to the basics, to what they can and can't do. Just small awareness sessions. Experts can come in for one hour to train employees and create that awareness around what AI can do, what it can't do, and what the side effects are. OpenCloud actually bought $4,000 worth of courses just to help the owner understand how business works.

Maya: Was that a mistake or intentional?

Dr. Adeel Shaikh Muhammad: It really did happen, and apparently there was no refund policy on those courses either.

Maya: That's so funny. And do you feel there is enough proper training going on currently inside the organisations you're encountering?

Dr. Adeel Shaikh Muhammad: No, obviously not. That's why shadow AI exists. People are using different AI tools, and I know some organisations very close to me where employees are using other AI tools via VPNs even though they're not permitted. It's like oxygen - they can't live without those tools.

Maya: That's fascinating. Is it a case of organisations, because they are large and risk-averse, moving relatively slowly on some of this - while the individuals, particularly the early adopters, know their jobs can be done more quickly and are therefore using other tools? Is that the dilemma?

Dr. Adeel Shaikh Muhammad: It's both. They're using AI for personal stuff, and sometimes they are literally causing data loss prevention issues - DLP. There was a study suggesting that around 60 to 70% of all contracts are being copy-pasted into these LLMs. Just imagine. That's a big number to digest, because we don't always understand the fine details of legal agreements and contracts, so we copy-paste them into ChatGPT and ask what needs to be flagged or changed.

Maya: So there's a real push and pull here. The organisation needs to protect itself, but at the same time individuals want to extract value from the right tools and know their jobs can be done more quickly. That tension needs to be addressed continuously. Good training can help, but there also needs to be more widespread leadership-level management of this.

Dr. Adeel Shaikh Muhammad: There are also simulations that can help. Like, an employee might receive a simulated email prompting them to upload something, and then be shown the consequences of what could happen. There are some simulation tools and gamified approaches around this. But again, you need to bring in professionals who can do this for your organisation.

Where to find Dr. Adeel Shaikh Muhammad

Maya: We've had two fascinating conversations and you've clearly got so much knowledge to share. If people want to find out more about what you do and what you're saying about these topics, where's the best place to find you?

Dr. Adeel Shaikh Muhammad: They can find me on LinkedIn, or they can go to adeel.solutions for more information.

Maya: And do you share content on LinkedIn?

Dr. Adeel Shaikh Muhammad: Not too much, but I try. The best place is still adeel.solutions - that's my website, showcasing Adeel Solutions LLC, my legal entity. All the information is there, and you can be redirected to LinkedIn from there as well.

Maya: Brilliant. So people can see what you have to offer in terms of the ways you're working with teams at present. Thank you so much - I have learnt loads and really enjoyed this conversation.

Dr. Adeel Shaikh Muhammad: Thank you so very much, Maya, for giving me this platform. This is one of the greatest podcasts I have ever listened to. Really, really awesome, and thanks so much for your time.

Maya: And thank you all for listening. If you know somebody who has been struggling with these ideas or wants to know more, please send them a link to these episodes. It costs nothing, and it could completely transform their productivity over the next few years. I look forward to connecting with you next time.

You can connect with Dr Adeel Shaikh Muhammad at https://www.linkedin.com/in/shadeel/

Enjoyed this episode? Catch up here:

Watch on YouTube | Listen on Apple Podcasts | Spotify

About Maya

Maya Gudka is an executive coach specialising in C-suite career progression and leadership development. She works with senior leaders in major organisations on strategic career planning, executive presence, and building sustainable influence. Maya hosts The Refreshing Leadership Podcast, which ranks in the top 2% of podcasts globally and has nearly 300 episodes exploring the challenges faced by ambitious professionals.